feat(agent): Add EXIT_ON_DNS_ERROR environment variable (#1929)

Co-authored-by: henrygd <hank@henrygd.me>

.github/workflows/inactivity-actions.yml

@@ -6,6 +6,7 @@ on:
   workflow_dispatch:
 
 permissions:
+  actions: write
   issues: write
   pull-requests: write
 
@@ -48,6 +49,9 @@ jobs:
           # Action can not skip PRs, set it to 100 years to cover it.
           days-before-pr-stale: 36524
 
+          # Max issues to process before early exit. Next run resumes from cache. GH API limit: 5000.
+          operations-per-run: 1500
+
           # Labels
           stale-issue-label: 'stale'
           remove-stale-when-updated: true
@@ -56,4 +60,5 @@ jobs:
 
           # Exemptions
           exempt-assignees: true
-          exempt-milestones: true
+
+          exempt-milestones: true

.github/workflows/vulncheck.yml

@@ -19,11 +19,11 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Check out code into the Go module directory
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
       - name: Set up Go
-        uses: actions/setup-go@v5
+        uses: actions/setup-go@v6
         with:
-          go-version: 1.25.x
+          go-version: 1.26.x
           # cached: false
       - name: Get official govulncheck
         run: go install golang.org/x/vuln/cmd/govulncheck@latest

.gitignore

@@ -10,6 +10,7 @@ dist
 *.exe
 internal/cmd/hub/hub
 internal/cmd/agent/agent
+agent.test
 node_modules
 build
 *timestamp*

Makefile

@@ -3,6 +3,40 @@ OS ?= $(shell go env GOOS)
 ARCH ?= $(shell go env GOARCH)
 # Skip building the web UI if true
 SKIP_WEB ?= false
+# Controls NVML/glibc agent build tag behavior:
+# - auto (default): enable on linux/amd64 glibc hosts
+# - true: always enable
+# - false: always disable
+NVML ?= auto
+
+# Detect glibc host for local linux/amd64 builds.
+HOST_GLIBC := $(shell \
+	if [ "$(OS)" = "linux" ] && [ "$(ARCH)" = "amd64" ]; then \
+		for p in /lib64/ld-linux-x86-64.so.2 /lib/x86_64-linux-gnu/ld-linux-x86-64.so.2 /lib/ld-linux-x86-64.so.2; do \
+			[ -e "$$p" ] && { echo true; exit 0; }; \
+		done; \
+		if command -v ldd >/dev/null 2>&1; then \
+			if ldd --version 2>&1 | tr '[:upper:]' '[:lower:]' | awk '/gnu libc|glibc/{found=1} END{exit !found}'; then \
+				echo true; \
+			else \
+				echo false; \
+			fi; \
+		else \
+			echo false; \
+		fi; \
+	else \
+		echo false; \
+	fi)
+
+# Enable glibc build tag for NVML on supported Linux builds.
+AGENT_GO_TAGS :=
+ifeq ($(NVML),true)
+AGENT_GO_TAGS := -tags glibc
+else ifeq ($(NVML),auto)
+ifeq ($(HOST_GLIBC),true)
+AGENT_GO_TAGS := -tags glibc
+endif
+endif
 
 # Set executable extension based on target OS
 EXE_EXT := $(if $(filter windows,$(OS)),.exe,)
@@ -17,7 +51,6 @@ clean:
 lint:
 	golangci-lint run
 
-test: export GOEXPERIMENT=synctest
 test:
 	go test -tags=testing ./...
 
@@ -54,7 +87,7 @@ fetch-smartctl-conditional:
 
 # Update build-agent to include conditional .NET build
 build-agent: tidy build-dotnet-conditional fetch-smartctl-conditional
-	GOOS=$(OS) GOARCH=$(ARCH) go build -o ./build/beszel-agent_$(OS)_$(ARCH)$(EXE_EXT) -ldflags "-w -s" ./internal/cmd/agent
+	GOOS=$(OS) GOARCH=$(ARCH) go build $(AGENT_GO_TAGS) -o ./build/beszel-agent_$(OS)_$(ARCH)$(EXE_EXT) -ldflags "-w -s" ./internal/cmd/agent
 
 build-hub: tidy $(if $(filter false,$(SKIP_WEB)),build-web-ui)
 	GOOS=$(OS) GOARCH=$(ARCH) go build -o ./build/beszel_$(OS)_$(ARCH)$(EXE_EXT) -ldflags "-w -s" ./internal/cmd/hub
@@ -90,9 +123,9 @@ dev-hub:
 
 dev-agent:
 	@if command -v entr >/dev/null 2>&1; then \
-		find ./internal/cmd/agent/*.go ./agent/*.go | entr -r go run github.com/henrygd/beszel/internal/cmd/agent; \
+		find ./internal/cmd/agent/*.go ./agent/*.go | entr -r go run $(AGENT_GO_TAGS) github.com/henrygd/beszel/internal/cmd/agent; \
 	else \
-		go run github.com/henrygd/beszel/internal/cmd/agent; \
+		go run $(AGENT_GO_TAGS) github.com/henrygd/beszel/internal/cmd/agent; \
 	fi
 	
 build-dotnet:

agent/agent.go

@@ -5,11 +5,7 @@
 package agent
 
 import (
-	"crypto/sha256"
-	"encoding/hex"
 	"log/slog"
-	"os"
-	"path/filepath"
 	"strings"
 	"sync"
 	"time"
@@ -17,12 +13,14 @@ import (
 	"github.com/gliderlabs/ssh"
 	"github.com/henrygd/beszel"
 	"github.com/henrygd/beszel/agent/deltatracker"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/common"
 	"github.com/henrygd/beszel/internal/entities/system"
-	"github.com/shirou/gopsutil/v4/host"
 	gossh "golang.org/x/crypto/ssh"
 )
 
+const defaultDataCacheTimeMs uint16 = 60_000
+
 type Agent struct {
 	sync.Mutex                                                                      // Used to lock agent while collecting data
 	debug                     bool                                                  // true if LOG_LEVEL is set to debug
@@ -40,6 +38,7 @@ type Agent struct {
 	sensorConfig              *SensorConfig                                         // Sensors config
 	systemInfo                system.Info                                           // Host system info (dynamic)
 	systemDetails             system.Details                                        // Host system details (static, once-per-connection)
+	detailsDirty              bool                                                  // Whether system details have changed and need to be resent
 	gpuManager                *GPUManager                                           // Manages GPU data
 	cache                     *systemDataCache                                      // Cache for system stats based on cache time
 	connectionManager         *ConnectionManager                                    // Channel to signal connection events
@@ -65,18 +64,18 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
 	agent.netIoStats = make(map[uint16]system.NetIoStats)
 	agent.netInterfaceDeltaTrackers = make(map[uint16]*deltatracker.DeltaTracker[string, uint64])
 
-	agent.dataDir, err = getDataDir(dataDir...)
+	agent.dataDir, err = GetDataDir(dataDir...)
 	if err != nil {
 		slog.Warn("Data directory not found")
 	} else {
 		slog.Info("Data directory", "path", agent.dataDir)
 	}
 
-	agent.memCalc, _ = GetEnv("MEM_CALC")
+	agent.memCalc, _ = utils.GetEnv("MEM_CALC")
 	agent.sensorConfig = agent.newSensorConfig()
 
 	// Parse disk usage cache duration (e.g., "15m", "1h") to avoid waking sleeping disks
-	if diskUsageCache, exists := GetEnv("DISK_USAGE_CACHE"); exists {
+	if diskUsageCache, exists := utils.GetEnv("DISK_USAGE_CACHE"); exists {
 		if duration, err := time.ParseDuration(diskUsageCache); err == nil {
 			agent.diskUsageCacheDuration = duration
 			slog.Info("DISK_USAGE_CACHE", "duration", duration)
@@ -86,7 +85,7 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
 	}
 
 	// Set up slog with a log level determined by the LOG_LEVEL env var
-	if logLevelStr, exists := GetEnv("LOG_LEVEL"); exists {
+	if logLevelStr, exists := utils.GetEnv("LOG_LEVEL"); exists {
 		switch strings.ToLower(logLevelStr) {
 		case "debug":
 			agent.debug = true
@@ -101,13 +100,13 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
 	slog.Debug(beszel.Version)
 
 	// initialize docker manager
-	agent.dockerManager = newDockerManager()
+	agent.dockerManager = newDockerManager(agent)
 
 	// initialize system info
 	agent.refreshSystemDetails()
 
 	// SMART_INTERVAL env var to update smart data at this interval
-	if smartIntervalEnv, exists := GetEnv("SMART_INTERVAL"); exists {
+	if smartIntervalEnv, exists := utils.GetEnv("SMART_INTERVAL"); exists {
 		if duration, err := time.ParseDuration(smartIntervalEnv); err == nil && duration > 0 {
 			agent.systemDetails.SmartInterval = duration
 			slog.Info("SMART_INTERVAL", "duration", duration)
@@ -146,21 +145,12 @@ func NewAgent(dataDir ...string) (agent *Agent, err error) {
 
 	// if debugging, print stats
 	if agent.debug {
-		slog.Debug("Stats", "data", agent.gatherStats(common.DataRequestOptions{CacheTimeMs: 60_000, IncludeDetails: true}))
+		slog.Debug("Stats", "data", agent.gatherStats(common.DataRequestOptions{CacheTimeMs: defaultDataCacheTimeMs, IncludeDetails: true}))
 	}
 
 	return agent, nil
 }
 
-// GetEnv retrieves an environment variable with a "BESZEL_AGENT_" prefix, or falls back to the unprefixed key.
-func GetEnv(key string) (value string, exists bool) {
-	if value, exists = os.LookupEnv("BESZEL_AGENT_" + key); exists {
-		return value, exists
-	}
-	// Fallback to the old unprefixed key
-	return os.LookupEnv(key)
-}
-
 func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedData {
 	a.Lock()
 	defer a.Unlock()
@@ -177,11 +167,6 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
 		Info:  a.systemInfo,
 	}
 
-	// Include static system details only when requested
-	if options.IncludeDetails {
-		data.Details = &a.systemDetails
-	}
-
 	// slog.Info("System data", "data", data, "cacheTimeMs", cacheTimeMs)
 
 	if a.dockerManager != nil {
@@ -194,7 +179,7 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
 	}
 
 	// skip updating systemd services if cache time is not the default 60sec interval
-	if a.systemdManager != nil && cacheTimeMs == 60_000 {
+	if a.systemdManager != nil && cacheTimeMs == defaultDataCacheTimeMs {
 		totalCount := uint16(a.systemdManager.getServiceStatsCount())
 		if totalCount > 0 {
 			numFailed := a.systemdManager.getFailedServiceCount()
@@ -217,7 +202,7 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
 			data.Stats.ExtraFs[key] = stats
 			// Add percentages to Info struct for dashboard
 			if stats.DiskTotal > 0 {
-				pct := twoDecimals((stats.DiskUsed / stats.DiskTotal) * 100)
+				pct := utils.TwoDecimals((stats.DiskUsed / stats.DiskTotal) * 100)
 				data.Info.ExtraFsPct[key] = pct
 			}
 		}
@@ -225,43 +210,16 @@ func (a *Agent) gatherStats(options common.DataRequestOptions) *system.CombinedD
 	slog.Debug("Extra FS", "data", data.Stats.ExtraFs)
 
 	a.cache.Set(data, cacheTimeMs)
-	return data
+
+	return a.attachSystemDetails(data, cacheTimeMs, options.IncludeDetails)
 }
 
-// StartAgent initializes and starts the agent with optional WebSocket connection
+// Start initializes and starts the agent with optional WebSocket connection
 func (a *Agent) Start(serverOptions ServerOptions) error {
 	a.keys = serverOptions.Keys
 	return a.connectionManager.Start(serverOptions)
 }
 
 func (a *Agent) getFingerprint() string {
-	// first look for a fingerprint in the data directory
-	if a.dataDir != "" {
-		if fp, err := os.ReadFile(filepath.Join(a.dataDir, "fingerprint")); err == nil {
-			if s := strings.TrimSpace(string(fp)); s != "" {
-				return s
-			}
-		}
-	}
-
-	// if no fingerprint is found, generate one
-	fingerprint, err := host.HostID()
-	// we ignore a commonly known "product_uuid" known not to be unique
-	if err != nil || fingerprint == "" || fingerprint == "03000200-0400-0500-0006-000700080009" {
-		fingerprint = a.systemDetails.Hostname + a.systemDetails.CpuModel
-	}
-
-	// hash fingerprint
-	sum := sha256.Sum256([]byte(fingerprint))
-	fingerprint = hex.EncodeToString(sum[:24])
-
-	// save fingerprint to data directory
-	if a.dataDir != "" {
-		err = os.WriteFile(filepath.Join(a.dataDir, "fingerprint"), []byte(fingerprint), 0644)
-		if err != nil {
-			slog.Warn("Failed to save fingerprint", "err", err)
-		}
-	}
-
-	return fingerprint
+	return GetFingerprint(a.dataDir, a.systemDetails.Hostname, a.systemDetails.CpuModel)
 }

agent/agent_cache_test.go

@@ -1,5 +1,4 @@
 //go:build testing
-// +build testing
 
 package agent
 

agent/agent_test_helpers.go

@@ -1,5 +1,4 @@
 //go:build testing
-// +build testing
 
 package agent
 

agent/battery/battery.go

@@ -1,84 +1,11 @@
-//go:build !freebsd
-
-// Package battery provides functions to check if the system has a battery and to get the battery stats.
+// Package battery provides functions to check if the system has a battery and return the charge state and percentage.
 package battery
 
-import (
-	"errors"
-	"log/slog"
-	"math"
-
-	"github.com/distatus/battery"
+const (
+	stateUnknown uint8 = iota
+	stateEmpty
+	stateFull
+	stateCharging
+	stateDischarging
+	stateIdle
 )
-
-var (
-	systemHasBattery   = false
-	haveCheckedBattery = false
-)
-
-// HasReadableBattery checks if the system has a battery and returns true if it does.
-func HasReadableBattery() bool {
-	if haveCheckedBattery {
-		return systemHasBattery
-	}
-	haveCheckedBattery = true
-	batteries, err := battery.GetAll()
-	for _, bat := range batteries {
-		if bat != nil && (bat.Full > 0 || bat.Design > 0) {
-			systemHasBattery = true
-			break
-		}
-	}
-	if !systemHasBattery {
-		slog.Debug("No battery found", "err", err)
-	}
-	return systemHasBattery
-}
-
-// GetBatteryStats returns the current battery percent and charge state
-// percent = (current charge of all batteries) / (sum of designed/full capacity of all batteries)
-func GetBatteryStats() (batteryPercent uint8, batteryState uint8, err error) {
-	if !HasReadableBattery() {
-		return batteryPercent, batteryState, errors.ErrUnsupported
-	}
-	batteries, err := battery.GetAll()
-	// we'll handle errors later by skipping batteries with errors, rather
-	// than skipping everything because of the presence of some errors.
-	if len(batteries) == 0 {
-		return batteryPercent, batteryState, errors.New("no batteries")
-	}
-
-	totalCapacity := float64(0)
-	totalCharge := float64(0)
-	errs, partialErrs := err.(battery.Errors)
-
-	batteryState = math.MaxUint8
-
-	for i, bat := range batteries {
-		if partialErrs && errs[i] != nil {
-			// if there were some errors, like missing data, skip it
-			continue
-		}
-		if bat == nil || bat.Full == 0 {
-			// skip batteries with no capacity. Charge is unlikely to ever be zero, but
-			// we can't guarantee that, so don't skip based on charge.
-			continue
-		}
-		totalCapacity += bat.Full
-		totalCharge += min(bat.Current, bat.Full)
-		if bat.State.Raw >= 0 {
-			batteryState = uint8(bat.State.Raw)
-		}
-	}
-
-	if totalCapacity == 0 || batteryState == math.MaxUint8 {
-		// for macs there's sometimes a ghost battery with 0 capacity
-		// https://github.com/distatus/battery/issues/34
-		// Instead of skipping over those batteries, we'll check for total 0 capacity
-		// and return an error. This also prevents a divide by zero.
-		return batteryPercent, batteryState, errors.New("no battery capacity")
-	}
-
-	batteryPercent = uint8(totalCharge / totalCapacity * 100)
-	return batteryPercent, batteryState, nil
-}

agent/battery/battery_darwin.go

@@ -0,0 +1,96 @@
+//go:build darwin
+
+package battery
+
+import (
+	"errors"
+	"log/slog"
+	"math"
+	"os/exec"
+	"sync"
+
+	"howett.net/plist"
+)
+
+type macBattery struct {
+	CurrentCapacity   int  `plist:"CurrentCapacity"`
+	MaxCapacity       int  `plist:"MaxCapacity"`
+	FullyCharged      bool `plist:"FullyCharged"`
+	IsCharging        bool `plist:"IsCharging"`
+	ExternalConnected bool `plist:"ExternalConnected"`
+}
+
+func readMacBatteries() ([]macBattery, error) {
+	out, err := exec.Command("ioreg", "-n", "AppleSmartBattery", "-r", "-a").Output()
+	if err != nil {
+		return nil, err
+	}
+	if len(out) == 0 {
+		return nil, nil
+	}
+	var batteries []macBattery
+	if _, err := plist.Unmarshal(out, &batteries); err != nil {
+		return nil, err
+	}
+	return batteries, nil
+}
+
+// HasReadableBattery checks if the system has a battery and returns true if it does.
+var HasReadableBattery = sync.OnceValue(func() bool {
+	systemHasBattery := false
+	batteries, err := readMacBatteries()
+	slog.Debug("Batteries", "batteries", batteries, "err", err)
+	for _, bat := range batteries {
+		if bat.MaxCapacity > 0 {
+			systemHasBattery = true
+			break
+		}
+	}
+	return systemHasBattery
+})
+
+// GetBatteryStats returns the current battery percent and charge state.
+// Uses CurrentCapacity/MaxCapacity to match the value macOS displays.
+func GetBatteryStats() (batteryPercent uint8, batteryState uint8, err error) {
+	if !HasReadableBattery() {
+		return batteryPercent, batteryState, errors.ErrUnsupported
+	}
+	batteries, err := readMacBatteries()
+	if len(batteries) == 0 {
+		return batteryPercent, batteryState, errors.New("no batteries")
+	}
+
+	totalCapacity := 0
+	totalCharge := 0
+	batteryState = math.MaxUint8
+
+	for _, bat := range batteries {
+		if bat.MaxCapacity == 0 {
+			// skip ghost batteries with 0 capacity
+			// https://github.com/distatus/battery/issues/34
+			continue
+		}
+		totalCapacity += bat.MaxCapacity
+		totalCharge += min(bat.CurrentCapacity, bat.MaxCapacity)
+
+		switch {
+		case !bat.ExternalConnected:
+			batteryState = stateDischarging
+		case bat.IsCharging:
+			batteryState = stateCharging
+		case bat.CurrentCapacity == 0:
+			batteryState = stateEmpty
+		case !bat.FullyCharged:
+			batteryState = stateIdle
+		default:
+			batteryState = stateFull
+		}
+	}
+
+	if totalCapacity == 0 || batteryState == math.MaxUint8 {
+		return batteryPercent, batteryState, errors.New("no battery capacity")
+	}
+
+	batteryPercent = uint8(float64(totalCharge) / float64(totalCapacity) * 100)
+	return batteryPercent, batteryState, nil
+}

agent/battery/battery_linux.go

@@ -0,0 +1,120 @@
+//go:build linux
+
+package battery
+
+import (
+	"errors"
+	"log/slog"
+	"math"
+	"os"
+	"path/filepath"
+	"strconv"
+	"sync"
+
+	"github.com/henrygd/beszel/agent/utils"
+)
+
+// getBatteryPaths returns the paths of all batteries in /sys/class/power_supply
+var getBatteryPaths func() ([]string, error)
+
+// HasReadableBattery checks if the system has a battery and returns true if it does.
+var HasReadableBattery func() bool
+
+func init() {
+	resetBatteryState("/sys/class/power_supply")
+}
+
+// resetBatteryState resets the sync.Once functions to a fresh state.
+// Tests call this after swapping sysfsPowerSupply so the new path is picked up.
+func resetBatteryState(sysfsPowerSupplyPath string) {
+	getBatteryPaths = sync.OnceValues(func() ([]string, error) {
+		entries, err := os.ReadDir(sysfsPowerSupplyPath)
+		if err != nil {
+			return nil, err
+		}
+		var paths []string
+		for _, e := range entries {
+			path := filepath.Join(sysfsPowerSupplyPath, e.Name())
+			if utils.ReadStringFile(filepath.Join(path, "type")) == "Battery" {
+				paths = append(paths, path)
+			}
+		}
+		return paths, nil
+	})
+	HasReadableBattery = sync.OnceValue(func() bool {
+		systemHasBattery := false
+		paths, err := getBatteryPaths()
+		for _, path := range paths {
+			if _, ok := utils.ReadStringFileOK(filepath.Join(path, "capacity")); ok {
+				systemHasBattery = true
+				break
+			}
+		}
+		if !systemHasBattery {
+			slog.Debug("No battery found", "err", err)
+		}
+		return systemHasBattery
+	})
+}
+
+func parseSysfsState(status string) uint8 {
+	switch status {
+	case "Empty":
+		return stateEmpty
+	case "Full":
+		return stateFull
+	case "Charging":
+		return stateCharging
+	case "Discharging":
+		return stateDischarging
+	case "Not charging":
+		return stateIdle
+	default:
+		return stateUnknown
+	}
+}
+
+// GetBatteryStats returns the current battery percent and charge state.
+// Reads /sys/class/power_supply/*/capacity directly so the kernel-reported
+// value is used, which is always 0-100 and matches what the OS displays.
+func GetBatteryStats() (batteryPercent uint8, batteryState uint8, err error) {
+	if !HasReadableBattery() {
+		return batteryPercent, batteryState, errors.ErrUnsupported
+	}
+	paths, err := getBatteryPaths()
+	if err != nil {
+		return batteryPercent, batteryState, err
+	}
+	if len(paths) == 0 {
+		return batteryPercent, batteryState, errors.New("no batteries")
+	}
+
+	batteryState = math.MaxUint8
+	totalPercent := 0
+	count := 0
+
+	for _, path := range paths {
+		capStr, ok := utils.ReadStringFileOK(filepath.Join(path, "capacity"))
+		if !ok {
+			continue
+		}
+		cap, parseErr := strconv.Atoi(capStr)
+		if parseErr != nil {
+			continue
+		}
+		totalPercent += cap
+		count++
+
+		state := parseSysfsState(utils.ReadStringFile(filepath.Join(path, "status")))
+		if state != stateUnknown {
+			batteryState = state
+		}
+	}
+
+	if count == 0 || batteryState == math.MaxUint8 {
+		return batteryPercent, batteryState, errors.New("no battery capacity")
+	}
+
+	batteryPercent = uint8(totalPercent / count)
+	return batteryPercent, batteryState, nil
+}

agent/battery/battery_linux_test.go

@@ -0,0 +1,201 @@
+//go:build testing && linux
+
+package battery
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+// setupFakeSysfs creates a temporary sysfs-like tree under t.TempDir(),
+// swaps sysfsPowerSupply, resets the sync.Once caches, and restores
+// everything on cleanup. Returns a helper to create battery directories.
+func setupFakeSysfs(t *testing.T) (tmpDir string, addBattery func(name, capacity, status string)) {
+	t.Helper()
+
+	tmp := t.TempDir()
+	resetBatteryState(tmp)
+
+	write := func(path, content string) {
+		t.Helper()
+		dir := filepath.Dir(path)
+		if err := os.MkdirAll(dir, 0o755); err != nil {
+			t.Fatal(err)
+		}
+		if err := os.WriteFile(path, []byte(content), 0o644); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	addBattery = func(name, capacity, status string) {
+		t.Helper()
+		batDir := filepath.Join(tmp, name)
+		write(filepath.Join(batDir, "type"), "Battery")
+		write(filepath.Join(batDir, "capacity"), capacity)
+		write(filepath.Join(batDir, "status"), status)
+	}
+
+	return tmp, addBattery
+}
+
+func TestParseSysfsState(t *testing.T) {
+	tests := []struct {
+		input string
+		want  uint8
+	}{
+		{"Empty", stateEmpty},
+		{"Full", stateFull},
+		{"Charging", stateCharging},
+		{"Discharging", stateDischarging},
+		{"Not charging", stateIdle},
+		{"", stateUnknown},
+		{"SomethingElse", stateUnknown},
+	}
+	for _, tt := range tests {
+		assert.Equal(t, tt.want, parseSysfsState(tt.input), "parseSysfsState(%q)", tt.input)
+	}
+}
+
+func TestGetBatteryStats_SingleBattery(t *testing.T) {
+	_, addBattery := setupFakeSysfs(t)
+	addBattery("BAT0", "72", "Discharging")
+
+	pct, state, err := GetBatteryStats()
+	assert.NoError(t, err)
+	assert.Equal(t, uint8(72), pct)
+	assert.Equal(t, stateDischarging, state)
+}
+
+func TestGetBatteryStats_MultipleBatteries(t *testing.T) {
+	_, addBattery := setupFakeSysfs(t)
+	addBattery("BAT0", "80", "Charging")
+	addBattery("BAT1", "40", "Charging")
+
+	pct, state, err := GetBatteryStats()
+	assert.NoError(t, err)
+	// average of 80 and 40 = 60
+	assert.EqualValues(t, 60, pct)
+	assert.Equal(t, stateCharging, state)
+}
+
+func TestGetBatteryStats_FullBattery(t *testing.T) {
+	_, addBattery := setupFakeSysfs(t)
+	addBattery("BAT0", "100", "Full")
+
+	pct, state, err := GetBatteryStats()
+	assert.NoError(t, err)
+	assert.Equal(t, uint8(100), pct)
+	assert.Equal(t, stateFull, state)
+}
+
+func TestGetBatteryStats_EmptyBattery(t *testing.T) {
+	_, addBattery := setupFakeSysfs(t)
+	addBattery("BAT0", "0", "Empty")
+
+	pct, state, err := GetBatteryStats()
+	assert.NoError(t, err)
+	assert.Equal(t, uint8(0), pct)
+	assert.Equal(t, stateEmpty, state)
+}
+
+func TestGetBatteryStats_NotCharging(t *testing.T) {
+	_, addBattery := setupFakeSysfs(t)
+	addBattery("BAT0", "80", "Not charging")
+
+	pct, state, err := GetBatteryStats()
+	assert.NoError(t, err)
+	assert.Equal(t, uint8(80), pct)
+	assert.Equal(t, stateIdle, state)
+}
+
+func TestGetBatteryStats_NoBatteries(t *testing.T) {
+	setupFakeSysfs(t) // empty directory, no batteries
+
+	_, _, err := GetBatteryStats()
+	assert.Error(t, err)
+}
+
+func TestGetBatteryStats_NonBatterySupplyIgnored(t *testing.T) {
+	tmp, addBattery := setupFakeSysfs(t)
+
+	// Add a real battery
+	addBattery("BAT0", "55", "Charging")
+
+	// Add an AC adapter (type != Battery) - should be ignored
+	acDir := filepath.Join(tmp, "AC0")
+	if err := os.MkdirAll(acDir, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(acDir, "type"), []byte("Mains"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	pct, state, err := GetBatteryStats()
+	assert.NoError(t, err)
+	assert.Equal(t, uint8(55), pct)
+	assert.Equal(t, stateCharging, state)
+}
+
+func TestGetBatteryStats_InvalidCapacitySkipped(t *testing.T) {
+	tmp, addBattery := setupFakeSysfs(t)
+
+	// One battery with valid capacity
+	addBattery("BAT0", "90", "Discharging")
+
+	// Another with invalid capacity text
+	badDir := filepath.Join(tmp, "BAT1")
+	if err := os.MkdirAll(badDir, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(badDir, "type"), []byte("Battery"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(badDir, "capacity"), []byte("not-a-number"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(badDir, "status"), []byte("Discharging"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	pct, _, err := GetBatteryStats()
+	assert.NoError(t, err)
+	// Only BAT0 counted
+	assert.Equal(t, uint8(90), pct)
+}
+
+func TestGetBatteryStats_UnknownStatusOnly(t *testing.T) {
+	_, addBattery := setupFakeSysfs(t)
+	addBattery("BAT0", "50", "SomethingWeird")
+
+	_, _, err := GetBatteryStats()
+	assert.Error(t, err)
+}
+
+func TestHasReadableBattery_True(t *testing.T) {
+	_, addBattery := setupFakeSysfs(t)
+	addBattery("BAT0", "50", "Charging")
+
+	assert.True(t, HasReadableBattery())
+}
+
+func TestHasReadableBattery_False(t *testing.T) {
+	setupFakeSysfs(t) // no batteries
+
+	assert.False(t, HasReadableBattery())
+}
+
+func TestHasReadableBattery_NoCapacityFile(t *testing.T) {
+	tmp, _ := setupFakeSysfs(t)
+
+	// Battery dir with type file but no capacity file
+	batDir := filepath.Join(tmp, "BAT0")
+	err := os.MkdirAll(batDir, 0o755)
+	assert.NoError(t, err)
+	err = os.WriteFile(filepath.Join(batDir, "type"), []byte("Battery"), 0o644)
+	assert.NoError(t, err)
+
+	assert.False(t, HasReadableBattery())
+}

agent/battery/battery_stub.go

@@ -1,4 +1,4 @@
-//go:build freebsd
+//go:build !darwin && !linux && !windows
 
 package battery
 

agent/battery/battery_windows.go

@@ -0,0 +1,298 @@
+//go:build windows
+
+// Most of the Windows battery code is based on
+// distatus/battery by Karol 'Kenji Takahashi' Woźniak
+
+package battery
+
+import (
+	"errors"
+	"log/slog"
+	"math"
+	"sync"
+	"syscall"
+	"unsafe"
+
+	"golang.org/x/sys/windows"
+)
+
+type batteryQueryInformation struct {
+	BatteryTag       uint32
+	InformationLevel int32
+	AtRate           int32
+}
+
+type batteryInformation struct {
+	Capabilities        uint32
+	Technology          uint8
+	Reserved            [3]uint8
+	Chemistry           [4]uint8
+	DesignedCapacity    uint32
+	FullChargedCapacity uint32
+	DefaultAlert1       uint32
+	DefaultAlert2       uint32
+	CriticalBias        uint32
+	CycleCount          uint32
+}
+
+type batteryWaitStatus struct {
+	BatteryTag   uint32
+	Timeout      uint32
+	PowerState   uint32
+	LowCapacity  uint32
+	HighCapacity uint32
+}
+
+type batteryStatus struct {
+	PowerState uint32
+	Capacity   uint32
+	Voltage    uint32
+	Rate       int32
+}
+
+type winGUID struct {
+	Data1 uint32
+	Data2 uint16
+	Data3 uint16
+	Data4 [8]byte
+}
+
+type spDeviceInterfaceData struct {
+	cbSize             uint32
+	InterfaceClassGuid winGUID
+	Flags              uint32
+	Reserved           uint
+}
+
+var guidDeviceBattery = winGUID{
+	0x72631e54,
+	0x78A4,
+	0x11d0,
+	[8]byte{0xbc, 0xf7, 0x00, 0xaa, 0x00, 0xb7, 0xb3, 0x2a},
+}
+
+var (
+	setupapi                         = &windows.LazyDLL{Name: "setupapi.dll", System: true}
+	setupDiGetClassDevsW             = setupapi.NewProc("SetupDiGetClassDevsW")
+	setupDiEnumDeviceInterfaces      = setupapi.NewProc("SetupDiEnumDeviceInterfaces")
+	setupDiGetDeviceInterfaceDetailW = setupapi.NewProc("SetupDiGetDeviceInterfaceDetailW")
+	setupDiDestroyDeviceInfoList     = setupapi.NewProc("SetupDiDestroyDeviceInfoList")
+)
+
+// winBatteryGet reads one battery by index. Returns (fullCapacity, currentCapacity, state, error).
+// Returns error == errNotFound when there are no more batteries.
+var errNotFound = errors.New("no more batteries")
+
+func setupDiSetup(proc *windows.LazyProc, nargs, a1, a2, a3, a4, a5, a6 uintptr) (uintptr, error) {
+	_ = nargs
+	r1, _, errno := syscall.SyscallN(proc.Addr(), a1, a2, a3, a4, a5, a6)
+	if windows.Handle(r1) == windows.InvalidHandle {
+		if errno != 0 {
+			return 0, error(errno)
+		}
+		return 0, syscall.EINVAL
+	}
+	return r1, nil
+}
+
+func setupDiCall(proc *windows.LazyProc, nargs, a1, a2, a3, a4, a5, a6 uintptr) syscall.Errno {
+	_ = nargs
+	r1, _, errno := syscall.SyscallN(proc.Addr(), a1, a2, a3, a4, a5, a6)
+	if r1 == 0 {
+		if errno != 0 {
+			return errno
+		}
+		return syscall.EINVAL
+	}
+	return 0
+}
+
+func readWinBatteryState(powerState uint32) uint8 {
+	switch {
+	case powerState&0x00000004 != 0:
+		return stateCharging
+	case powerState&0x00000008 != 0:
+		return stateEmpty
+	case powerState&0x00000002 != 0:
+		return stateDischarging
+	case powerState&0x00000001 != 0:
+		return stateFull
+	default:
+		return stateUnknown
+	}
+}
+
+func winBatteryGet(idx int) (full, current uint32, state uint8, err error) {
+	hdev, err := setupDiSetup(
+		setupDiGetClassDevsW,
+		4,
+		uintptr(unsafe.Pointer(&guidDeviceBattery)),
+		0, 0,
+		2|16, // DIGCF_PRESENT|DIGCF_DEVICEINTERFACE
+		0, 0,
+	)
+	if err != nil {
+		return 0, 0, stateUnknown, err
+	}
+	defer syscall.SyscallN(setupDiDestroyDeviceInfoList.Addr(), hdev)
+
+	var did spDeviceInterfaceData
+	did.cbSize = uint32(unsafe.Sizeof(did))
+	errno := setupDiCall(
+		setupDiEnumDeviceInterfaces,
+		5,
+		hdev, 0,
+		uintptr(unsafe.Pointer(&guidDeviceBattery)),
+		uintptr(idx),
+		uintptr(unsafe.Pointer(&did)),
+		0,
+	)
+	if errno == 259 { // ERROR_NO_MORE_ITEMS
+		return 0, 0, stateUnknown, errNotFound
+	}
+	if errno != 0 {
+		return 0, 0, stateUnknown, errno
+	}
+
+	var cbRequired uint32
+	errno = setupDiCall(
+		setupDiGetDeviceInterfaceDetailW,
+		6,
+		hdev,
+		uintptr(unsafe.Pointer(&did)),
+		0, 0,
+		uintptr(unsafe.Pointer(&cbRequired)),
+		0,
+	)
+	if errno != 0 && errno != 122 { // ERROR_INSUFFICIENT_BUFFER
+		return 0, 0, stateUnknown, errno
+	}
+	didd := make([]uint16, cbRequired/2)
+	cbSize := (*uint32)(unsafe.Pointer(&didd[0]))
+	if unsafe.Sizeof(uint(0)) == 8 {
+		*cbSize = 8
+	} else {
+		*cbSize = 6
+	}
+	errno = setupDiCall(
+		setupDiGetDeviceInterfaceDetailW,
+		6,
+		hdev,
+		uintptr(unsafe.Pointer(&did)),
+		uintptr(unsafe.Pointer(&didd[0])),
+		uintptr(cbRequired),
+		uintptr(unsafe.Pointer(&cbRequired)),
+		0,
+	)
+	if errno != 0 {
+		return 0, 0, stateUnknown, errno
+	}
+	devicePath := &didd[2:][0]
+
+	handle, err := windows.CreateFile(
+		devicePath,
+		windows.GENERIC_READ|windows.GENERIC_WRITE,
+		windows.FILE_SHARE_READ|windows.FILE_SHARE_WRITE,
+		nil,
+		windows.OPEN_EXISTING,
+		windows.FILE_ATTRIBUTE_NORMAL,
+		0,
+	)
+	if err != nil {
+		return 0, 0, stateUnknown, err
+	}
+	defer windows.CloseHandle(handle)
+
+	var dwOut uint32
+	var dwWait uint32
+	var bqi batteryQueryInformation
+	err = windows.DeviceIoControl(
+		handle,
+		2703424, // IOCTL_BATTERY_QUERY_TAG
+		(*byte)(unsafe.Pointer(&dwWait)),
+		uint32(unsafe.Sizeof(dwWait)),
+		(*byte)(unsafe.Pointer(&bqi.BatteryTag)),
+		uint32(unsafe.Sizeof(bqi.BatteryTag)),
+		&dwOut, nil,
+	)
+	if err != nil || bqi.BatteryTag == 0 {
+		return 0, 0, stateUnknown, errors.New("battery tag not returned")
+	}
+
+	var bi batteryInformation
+	if err = windows.DeviceIoControl(
+		handle,
+		2703428, // IOCTL_BATTERY_QUERY_INFORMATION
+		(*byte)(unsafe.Pointer(&bqi)),
+		uint32(unsafe.Sizeof(bqi)),
+		(*byte)(unsafe.Pointer(&bi)),
+		uint32(unsafe.Sizeof(bi)),
+		&dwOut, nil,
+	); err != nil {
+		return 0, 0, stateUnknown, err
+	}
+
+	bws := batteryWaitStatus{BatteryTag: bqi.BatteryTag}
+	var bs batteryStatus
+	if err = windows.DeviceIoControl(
+		handle,
+		2703436, // IOCTL_BATTERY_QUERY_STATUS
+		(*byte)(unsafe.Pointer(&bws)),
+		uint32(unsafe.Sizeof(bws)),
+		(*byte)(unsafe.Pointer(&bs)),
+		uint32(unsafe.Sizeof(bs)),
+		&dwOut, nil,
+	); err != nil {
+		return 0, 0, stateUnknown, err
+	}
+
+	if bs.Capacity == 0xffffffff { // BATTERY_UNKNOWN_CAPACITY
+		return 0, 0, stateUnknown, errors.New("battery capacity unknown")
+	}
+
+	return bi.FullChargedCapacity, bs.Capacity, readWinBatteryState(bs.PowerState), nil
+}
+
+// HasReadableBattery checks if the system has a battery and returns true if it does.
+var HasReadableBattery = sync.OnceValue(func() bool {
+	systemHasBattery := false
+	full, _, _, err := winBatteryGet(0)
+	if err == nil && full > 0 {
+		systemHasBattery = true
+	}
+	if !systemHasBattery {
+		slog.Debug("No battery found", "err", err)
+	}
+	return systemHasBattery
+})
+
+// GetBatteryStats returns the current battery percent and charge state.
+func GetBatteryStats() (batteryPercent uint8, batteryState uint8, err error) {
+	if !HasReadableBattery() {
+		return batteryPercent, batteryState, errors.ErrUnsupported
+	}
+
+	totalFull := uint32(0)
+	totalCurrent := uint32(0)
+	batteryState = math.MaxUint8
+
+	for i := 0; ; i++ {
+		full, current, state, bErr := winBatteryGet(i)
+		if errors.Is(bErr, errNotFound) {
+			break
+		}
+		if bErr != nil || full == 0 {
+			continue
+		}
+		totalFull += full
+		totalCurrent += min(current, full)
+		batteryState = state
+	}
+
+	if totalFull == 0 || batteryState == math.MaxUint8 {
+		return batteryPercent, batteryState, errors.New("no battery capacity")
+	}
+
+	batteryPercent = uint8(float64(totalCurrent) / float64(totalFull) * 100)
+	return batteryPercent, batteryState, nil
+}

agent/client.go

@@ -14,11 +14,13 @@ import (
 	"time"
 
 	"github.com/henrygd/beszel"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/common"
 
 	"github.com/fxamacker/cbor/v2"
 	"github.com/lxzan/gws"
 	"golang.org/x/crypto/ssh"
+	"golang.org/x/net/proxy"
 )
 
 const (
@@ -43,7 +45,7 @@ type WebSocketClient struct {
 // newWebSocketClient creates a new WebSocket client for the given agent.
 // It reads configuration from environment variables and validates the hub URL.
 func newWebSocketClient(agent *Agent) (client *WebSocketClient, err error) {
-	hubURLStr, exists := GetEnv("HUB_URL")
+	hubURLStr, exists := utils.GetEnv("HUB_URL")
 	if !exists {
 		return nil, errors.New("HUB_URL environment variable not set")
 	}
@@ -72,12 +74,12 @@ func newWebSocketClient(agent *Agent) (client *WebSocketClient, err error) {
 // If neither is set, it returns an error.
 func getToken() (string, error) {
 	// get token from env var
-	token, _ := GetEnv("TOKEN")
+	token, _ := utils.GetEnv("TOKEN")
 	if token != "" {
 		return token, nil
 	}
 	// get token from file
-	tokenFile, _ := GetEnv("TOKEN_FILE")
+	tokenFile, _ := utils.GetEnv("TOKEN_FILE")
 	if tokenFile == "" {
 		return "", errors.New("must set TOKEN or TOKEN_FILE")
 	}
@@ -103,6 +105,11 @@ func (client *WebSocketClient) getOptions() *gws.ClientOption {
 	}
 	client.hubURL.Path = path.Join(client.hubURL.Path, "api/beszel/agent-connect")
 
+	// make sure BESZEL_AGENT_ALL_PROXY works (GWS only checks ALL_PROXY)
+	if val := os.Getenv("BESZEL_AGENT_ALL_PROXY"); val != "" {
+		os.Setenv("ALL_PROXY", val)
+	}
+
 	client.options = &gws.ClientOption{
 		Addr:      client.hubURL.String(),
 		TlsConfig: &tls.Config{InsecureSkipVerify: true},
@@ -111,6 +118,9 @@ func (client *WebSocketClient) getOptions() *gws.ClientOption {
 			"X-Token":    []string{client.token},
 			"X-Beszel":   []string{beszel.Version},
 		},
+		NewDialer: func() (gws.Dialer, error) {
+			return proxy.FromEnvironment(), nil
+		},
 	}
 	return client.options
 }
@@ -197,7 +207,7 @@ func (client *WebSocketClient) handleAuthChallenge(msg *common.HubRequest[cbor.R
 	}
 
 	if authRequest.NeedSysInfo {
-		response.Name, _ = GetEnv("SYSTEM_NAME")
+		response.Name, _ = utils.GetEnv("SYSTEM_NAME")
 		response.Hostname = client.agent.systemDetails.Hostname
 		serverAddr := client.agent.connectionManager.serverOptions.Addr
 		_, response.Port, _ = net.SplitHostPort(serverAddr)

agent/client_test.go

@@ -1,5 +1,4 @@
 //go:build testing
-// +build testing
 
 package agent
 
@@ -71,19 +70,11 @@ func TestNewWebSocketClient(t *testing.T) {
 		t.Run(tc.name, func(t *testing.T) {
 			// Set up environment
 			if tc.hubURL != "" {
-				os.Setenv("BESZEL_AGENT_HUB_URL", tc.hubURL)
-			} else {
-				os.Unsetenv("BESZEL_AGENT_HUB_URL")
+				t.Setenv("BESZEL_AGENT_HUB_URL", tc.hubURL)
 			}
 			if tc.token != "" {
-				os.Setenv("BESZEL_AGENT_TOKEN", tc.token)
-			} else {
-				os.Unsetenv("BESZEL_AGENT_TOKEN")
+				t.Setenv("BESZEL_AGENT_TOKEN", tc.token)
 			}
-			defer func() {
-				os.Unsetenv("BESZEL_AGENT_HUB_URL")
-				os.Unsetenv("BESZEL_AGENT_TOKEN")
-			}()
 
 			client, err := newWebSocketClient(agent)
 
@@ -139,12 +130,8 @@ func TestWebSocketClient_GetOptions(t *testing.T) {
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {
 			// Set up environment
-			os.Setenv("BESZEL_AGENT_HUB_URL", tc.inputURL)
-			os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
-			defer func() {
-				os.Unsetenv("BESZEL_AGENT_HUB_URL")
-				os.Unsetenv("BESZEL_AGENT_TOKEN")
-			}()
+			t.Setenv("BESZEL_AGENT_HUB_URL", tc.inputURL)
+			t.Setenv("BESZEL_AGENT_TOKEN", "test-token")
 
 			client, err := newWebSocketClient(agent)
 			require.NoError(t, err)
@@ -186,12 +173,8 @@ func TestWebSocketClient_VerifySignature(t *testing.T) {
 	require.NoError(t, err)
 
 	// Set up environment
-	os.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080")
-	os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
-	defer func() {
-		os.Unsetenv("BESZEL_AGENT_HUB_URL")
-		os.Unsetenv("BESZEL_AGENT_TOKEN")
-	}()
+	t.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080")
+	t.Setenv("BESZEL_AGENT_TOKEN", "test-token")
 
 	client, err := newWebSocketClient(agent)
 	require.NoError(t, err)
@@ -259,12 +242,8 @@ func TestWebSocketClient_HandleHubRequest(t *testing.T) {
 	agent := createTestAgent(t)
 
 	// Set up environment
-	os.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080")
-	os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
-	defer func() {
-		os.Unsetenv("BESZEL_AGENT_HUB_URL")
-		os.Unsetenv("BESZEL_AGENT_TOKEN")
-	}()
+	t.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080")
+	t.Setenv("BESZEL_AGENT_TOKEN", "test-token")
 
 	client, err := newWebSocketClient(agent)
 	require.NoError(t, err)
@@ -351,13 +330,8 @@ func TestGetUserAgent(t *testing.T) {
 func TestWebSocketClient_Close(t *testing.T) {
 	agent := createTestAgent(t)
 
-	// Set up environment
-	os.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080")
-	os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
-	defer func() {
-		os.Unsetenv("BESZEL_AGENT_HUB_URL")
-		os.Unsetenv("BESZEL_AGENT_TOKEN")
-	}()
+	t.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080")
+	t.Setenv("BESZEL_AGENT_TOKEN", "test-token")
 
 	client, err := newWebSocketClient(agent)
 	require.NoError(t, err)
@@ -372,13 +346,8 @@ func TestWebSocketClient_Close(t *testing.T) {
 func TestWebSocketClient_ConnectRateLimit(t *testing.T) {
 	agent := createTestAgent(t)
 
-	// Set up environment
-	os.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080")
-	os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
-	defer func() {
-		os.Unsetenv("BESZEL_AGENT_HUB_URL")
-		os.Unsetenv("BESZEL_AGENT_TOKEN")
-	}()
+	t.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080")
+	t.Setenv("BESZEL_AGENT_TOKEN", "test-token")
 
 	client, err := newWebSocketClient(agent)
 	require.NoError(t, err)
@@ -394,20 +363,10 @@ func TestWebSocketClient_ConnectRateLimit(t *testing.T) {
 
 // TestGetToken tests the getToken function with various scenarios
 func TestGetToken(t *testing.T) {
-	unsetEnvVars := func() {
-		os.Unsetenv("BESZEL_AGENT_TOKEN")
-		os.Unsetenv("TOKEN")
-		os.Unsetenv("BESZEL_AGENT_TOKEN_FILE")
-		os.Unsetenv("TOKEN_FILE")
-	}
-
 	t.Run("token from TOKEN environment variable", func(t *testing.T) {
-		unsetEnvVars()
-
 		// Set TOKEN env var
 		expectedToken := "test-token-from-env"
-		os.Setenv("TOKEN", expectedToken)
-		defer os.Unsetenv("TOKEN")
+		t.Setenv("TOKEN", expectedToken)
 
 		token, err := getToken()
 		assert.NoError(t, err)
@@ -415,12 +374,9 @@ func TestGetToken(t *testing.T) {
 	})
 
 	t.Run("token from BESZEL_AGENT_TOKEN environment variable", func(t *testing.T) {
-		unsetEnvVars()
-
 		// Set BESZEL_AGENT_TOKEN env var (should take precedence)
 		expectedToken := "test-token-from-beszel-env"
-		os.Setenv("BESZEL_AGENT_TOKEN", expectedToken)
-		defer os.Unsetenv("BESZEL_AGENT_TOKEN")
+		t.Setenv("BESZEL_AGENT_TOKEN", expectedToken)
 
 		token, err := getToken()
 		assert.NoError(t, err)
@@ -428,8 +384,6 @@ func TestGetToken(t *testing.T) {
 	})
 
 	t.Run("token from TOKEN_FILE", func(t *testing.T) {
-		unsetEnvVars()
-
 		// Create a temporary token file
 		expectedToken := "test-token-from-file"
 		tokenFile, err := os.CreateTemp("", "token-test-*.txt")
@@ -441,8 +395,7 @@ func TestGetToken(t *testing.T) {
 		tokenFile.Close()
 
 		// Set TOKEN_FILE env var
-		os.Setenv("TOKEN_FILE", tokenFile.Name())
-		defer os.Unsetenv("TOKEN_FILE")
+		t.Setenv("TOKEN_FILE", tokenFile.Name())
 
 		token, err := getToken()
 		assert.NoError(t, err)
@@ -450,8 +403,6 @@ func TestGetToken(t *testing.T) {
 	})
 
 	t.Run("token from BESZEL_AGENT_TOKEN_FILE", func(t *testing.T) {
-		unsetEnvVars()
-
 		// Create a temporary token file
 		expectedToken := "test-token-from-beszel-file"
 		tokenFile, err := os.CreateTemp("", "token-test-*.txt")
@@ -463,8 +414,7 @@ func TestGetToken(t *testing.T) {
 		tokenFile.Close()
 
 		// Set BESZEL_AGENT_TOKEN_FILE env var (should take precedence)
-		os.Setenv("BESZEL_AGENT_TOKEN_FILE", tokenFile.Name())
-		defer os.Unsetenv("BESZEL_AGENT_TOKEN_FILE")
+		t.Setenv("BESZEL_AGENT_TOKEN_FILE", tokenFile.Name())
 
 		token, err := getToken()
 		assert.NoError(t, err)
@@ -472,8 +422,6 @@ func TestGetToken(t *testing.T) {
 	})
 
 	t.Run("TOKEN takes precedence over TOKEN_FILE", func(t *testing.T) {
-		unsetEnvVars()
-
 		// Create a temporary token file
 		fileToken := "token-from-file"
 		tokenFile, err := os.CreateTemp("", "token-test-*.txt")
@@ -486,12 +434,8 @@ func TestGetToken(t *testing.T) {
 
 		// Set both TOKEN and TOKEN_FILE
 		envToken := "token-from-env"
-		os.Setenv("TOKEN", envToken)
-		os.Setenv("TOKEN_FILE", tokenFile.Name())
-		defer func() {
-			os.Unsetenv("TOKEN")
-			os.Unsetenv("TOKEN_FILE")
-		}()
+		t.Setenv("TOKEN", envToken)
+		t.Setenv("TOKEN_FILE", tokenFile.Name())
 
 		token, err := getToken()
 		assert.NoError(t, err)
@@ -499,7 +443,10 @@ func TestGetToken(t *testing.T) {
 	})
 
 	t.Run("error when neither TOKEN nor TOKEN_FILE is set", func(t *testing.T) {
-		unsetEnvVars()
+		t.Setenv("BESZEL_AGENT_TOKEN", "")
+		t.Setenv("TOKEN", "")
+		t.Setenv("BESZEL_AGENT_TOKEN_FILE", "")
+		t.Setenv("TOKEN_FILE", "")
 
 		token, err := getToken()
 		assert.Error(t, err)
@@ -508,11 +455,8 @@ func TestGetToken(t *testing.T) {
 	})
 
 	t.Run("error when TOKEN_FILE points to non-existent file", func(t *testing.T) {
-		unsetEnvVars()
-
 		// Set TOKEN_FILE to a non-existent file
-		os.Setenv("TOKEN_FILE", "/non/existent/file.txt")
-		defer os.Unsetenv("TOKEN_FILE")
+		t.Setenv("TOKEN_FILE", "/non/existent/file.txt")
 
 		token, err := getToken()
 		assert.Error(t, err)
@@ -521,8 +465,6 @@ func TestGetToken(t *testing.T) {
 	})
 
 	t.Run("handles empty token file", func(t *testing.T) {
-		unsetEnvVars()
-
 		// Create an empty token file
 		tokenFile, err := os.CreateTemp("", "token-test-*.txt")
 		require.NoError(t, err)
@@ -530,8 +472,7 @@ func TestGetToken(t *testing.T) {
 		tokenFile.Close()
 
 		// Set TOKEN_FILE env var
-		os.Setenv("TOKEN_FILE", tokenFile.Name())
-		defer os.Unsetenv("TOKEN_FILE")
+		t.Setenv("TOKEN_FILE", tokenFile.Name())
 
 		token, err := getToken()
 		assert.NoError(t, err)
@@ -539,8 +480,6 @@ func TestGetToken(t *testing.T) {
 	})
 
 	t.Run("strips whitespace from TOKEN_FILE", func(t *testing.T) {
-		unsetEnvVars()
-
 		tokenWithWhitespace := "  test-token-with-whitespace  \n\t"
 		expectedToken := "test-token-with-whitespace"
 		tokenFile, err := os.CreateTemp("", "token-test-*.txt")
@@ -551,8 +490,7 @@ func TestGetToken(t *testing.T) {
 		require.NoError(t, err)
 		tokenFile.Close()
 
-		os.Setenv("TOKEN_FILE", tokenFile.Name())
-		defer os.Unsetenv("TOKEN_FILE")
+		t.Setenv("TOKEN_FILE", tokenFile.Name())
 
 		token, err := getToken()
 		assert.NoError(t, err)

agent/connection_manager.go

@@ -1,14 +1,18 @@
 package agent
 
 import (
+	"context"
 	"errors"
 	"log/slog"
+	"net"
 	"os"
 	"os/signal"
+	"strings"
 	"syscall"
 	"time"
 
 	"github.com/henrygd/beszel/agent/health"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"
 )
 
@@ -91,8 +95,8 @@ func (c *ConnectionManager) Start(serverOptions ServerOptions) error {
 	c.eventChan = make(chan ConnectionEvent, 1)
 
 	// signal handling for shutdown
-	sigChan := make(chan os.Signal, 1)
-	signal.Notify(sigChan, syscall.SIGINT, syscall.SIGTERM)
+	sigCtx, stopSignals := signal.NotifyContext(context.Background(), syscall.SIGINT, syscall.SIGTERM)
+	defer stopSignals()
 
 	c.startWsTicker()
 	c.connect()
@@ -109,15 +113,38 @@ func (c *ConnectionManager) Start(serverOptions ServerOptions) error {
 			_ = c.startWebSocketConnection()
 		case <-healthTicker:
 			_ = health.Update()
-		case <-sigChan:
-			slog.Info("Shutting down")
-			_ = c.agent.StopServer()
-			c.closeWebSocket()
-			return health.CleanUp()
+		case <-sigCtx.Done():
+			slog.Info("Shutting down", "cause", context.Cause(sigCtx))
+			return c.stop()
 		}
 	}
 }
 
+// stop does not stop the connection manager itself, just any active connections. The manager will attempt to reconnect after stopping, so this should only be called immediately before shutting down the entire agent.
+//
+// If we need or want to expose a graceful Stop method in the future, do something like this to actually stop the manager:
+//
+//	func (c *ConnectionManager) Start(serverOptions ServerOptions) error {
+//		ctx, cancel := context.WithCancel(context.Background())
+//		c.cancel = cancel
+//
+//		for {
+//			select {
+//			case <-ctx.Done():
+//				return c.stop()
+//			}
+//		}
+//	}
+//
+//	func (c *ConnectionManager) Stop() {
+//		c.cancel()
+//	}
+func (c *ConnectionManager) stop() error {
+	_ = c.agent.StopServer()
+	c.closeWebSocket()
+	return health.CleanUp()
+}
+
 // handleEvent processes connection events and updates the connection state accordingly.
 func (c *ConnectionManager) handleEvent(event ConnectionEvent) {
 	switch event {
@@ -185,9 +212,16 @@ func (c *ConnectionManager) connect() {
 
 	// Try WebSocket first, if it fails, start SSH server
 	err := c.startWebSocketConnection()
-	if err != nil && c.State == Disconnected {
-		c.startSSHServer()
-		c.startWsTicker()
+	if err != nil {
+		if shouldExitOnErr(err) {
+			time.Sleep(2 * time.Second) // prevent tight restart loop
+			_ = c.stop()
+			os.Exit(1)
+		}
+		if c.State == Disconnected {
+			c.startSSHServer()
+			c.startWsTicker()
+		}
 	}
 }
 
@@ -224,3 +258,14 @@ func (c *ConnectionManager) closeWebSocket() {
 		c.wsClient.Close()
 	}
 }
+
+// shouldExitOnErr checks if the error is a DNS resolution failure and if the
+// EXIT_ON_DNS_ERROR env var is set. https://github.com/henrygd/beszel/issues/1924.
+func shouldExitOnErr(err error) bool {
+	if val, _ := utils.GetEnv("EXIT_ON_DNS_ERROR"); val == "true" {
+		if opErr, ok := errors.AsType[*net.OpError](err); ok {
+			return strings.Contains(opErr.Err.Error(), "lookup")
+		}
+	}
+	return false
+}

agent/connection_manager_test.go

@@ -1,14 +1,13 @@
 //go:build testing
-// +build testing
 
 package agent
 
 import (
 	"crypto/ed25519"
+	"errors"
 	"fmt"
 	"net"
 	"net/url"
-	"os"
 	"testing"
 	"time"
 
@@ -184,10 +183,6 @@ func TestConnectionManager_TickerManagement(t *testing.T) {
 
 // TestConnectionManager_WebSocketConnectionFlow tests WebSocket connection logic
 func TestConnectionManager_WebSocketConnectionFlow(t *testing.T) {
-	if testing.Short() {
-		t.Skip("Skipping WebSocket connection test in short mode")
-	}
-
 	agent := createTestAgent(t)
 	cm := agent.connectionManager
 
@@ -197,19 +192,18 @@ func TestConnectionManager_WebSocketConnectionFlow(t *testing.T) {
 	assert.Equal(t, Disconnected, cm.State, "State should remain Disconnected after failed connection")
 
 	// Test with invalid URL
-	os.Setenv("BESZEL_AGENT_HUB_URL", "invalid-url")
-	os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
-	defer func() {
-		os.Unsetenv("BESZEL_AGENT_HUB_URL")
-		os.Unsetenv("BESZEL_AGENT_TOKEN")
-	}()
-
-	// Test with missing token
-	os.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080")
-	os.Unsetenv("BESZEL_AGENT_TOKEN")
+	t.Setenv("BESZEL_AGENT_HUB_URL", "1,33%")
+	t.Setenv("BESZEL_AGENT_TOKEN", "test-token")
 
 	_, err2 := newWebSocketClient(agent)
-	assert.Error(t, err2, "WebSocket client creation should fail without token")
+	assert.Error(t, err2, "WebSocket client creation should fail with invalid URL")
+
+	// Test with missing token
+	t.Setenv("BESZEL_AGENT_HUB_URL", "http://localhost:8080")
+	t.Setenv("BESZEL_AGENT_TOKEN", "")
+
+	_, err3 := newWebSocketClient(agent)
+	assert.Error(t, err3, "WebSocket client creation should fail without token")
 }
 
 // TestConnectionManager_ReconnectionLogic tests reconnection prevention logic
@@ -235,12 +229,8 @@ func TestConnectionManager_ConnectWithRateLimit(t *testing.T) {
 	cm := agent.connectionManager
 
 	// Set up environment for WebSocket client creation
-	os.Setenv("BESZEL_AGENT_HUB_URL", "ws://localhost:8080")
-	os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
-	defer func() {
-		os.Unsetenv("BESZEL_AGENT_HUB_URL")
-		os.Unsetenv("BESZEL_AGENT_TOKEN")
-	}()
+	t.Setenv("BESZEL_AGENT_HUB_URL", "ws://localhost:8080")
+	t.Setenv("BESZEL_AGENT_TOKEN", "test-token")
 
 	// Create WebSocket client
 	wsClient, err := newWebSocketClient(agent)
@@ -286,12 +276,8 @@ func TestConnectionManager_CloseWebSocket(t *testing.T) {
 	}, "Should not panic when closing nil WebSocket client")
 
 	// Set up environment and create WebSocket client
-	os.Setenv("BESZEL_AGENT_HUB_URL", "ws://localhost:8080")
-	os.Setenv("BESZEL_AGENT_TOKEN", "test-token")
-	defer func() {
-		os.Unsetenv("BESZEL_AGENT_HUB_URL")
-		os.Unsetenv("BESZEL_AGENT_TOKEN")
-	}()
+	t.Setenv("BESZEL_AGENT_HUB_URL", "ws://localhost:8080")
+	t.Setenv("BESZEL_AGENT_TOKEN", "test-token")
 
 	wsClient, err := newWebSocketClient(agent)
 	require.NoError(t, err)
@@ -313,3 +299,65 @@ func TestConnectionManager_ConnectFlow(t *testing.T) {
 		cm.connect()
 	}, "Connect should not panic without WebSocket client")
 }
+
+func TestShouldExitOnErr(t *testing.T) {
+	createDialErr := func(msg string) error {
+		return &net.OpError{
+			Op:  "dial",
+			Net: "tcp",
+			Err: errors.New(msg),
+		}
+	}
+
+	tests := []struct {
+		name     string
+		err      error
+		envValue string
+		expected bool
+	}{
+		{
+			name:     "no env var",
+			err:      createDialErr("lookup lkahsdfasdf: no such host"),
+			envValue: "",
+			expected: false,
+		},
+		{
+			name:     "env var false",
+			err:      createDialErr("lookup lkahsdfasdf: no such host"),
+			envValue: "false",
+			expected: false,
+		},
+		{
+			name:     "env var true, matching error",
+			err:      createDialErr("lookup lkahsdfasdf: no such host"),
+			envValue: "true",
+			expected: true,
+		},
+		{
+			name:     "env var true, matching error with extra context",
+			err:      createDialErr("lookup beszel.server.lan on [::1]:53: read udp [::1]:44557->[::1]:53: read: connection refused"),
+			envValue: "true",
+			expected: true,
+		},
+		{
+			name:     "env var true, non-matching error",
+			err:      errors.New("connection refused"),
+			envValue: "true",
+			expected: false,
+		},
+		{
+			name:     "env var true, dial but not lookup",
+			err:      createDialErr("connection timeout"),
+			envValue: "true",
+			expected: false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Setenv("EXIT_ON_DNS_ERROR", tt.envValue)
+			result := shouldExitOnErr(tt.err)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}

agent/cpu.go

@@ -14,10 +14,10 @@ var lastPerCoreCpuTimes = make(map[uint16][]cpu.TimesStat)
 // init initializes the CPU monitoring by storing the initial CPU times
 // for the default 60-second cache interval.
 func init() {
-	if times, err := cpu.Times(false); err == nil {
+	if times, err := cpu.Times(false); err == nil && len(times) > 0 {
 		lastCpuTimes[60000] = times[0]
 	}
-	if perCoreTimes, err := cpu.Times(true); err == nil {
+	if perCoreTimes, err := cpu.Times(true); err == nil && len(perCoreTimes) > 0 {
 		lastPerCoreCpuTimes[60000] = perCoreTimes
 	}
 }
@@ -89,10 +89,7 @@ func getPerCoreCpuUsage(cacheTimeMs uint16) (system.Uint8Slice, error) {
 	lastTimes := lastPerCoreCpuTimes[cacheTimeMs]
 
 	// Limit to the number of cores available in both samples
-	length := len(perCoreTimes)
-	if len(lastTimes) < length {
-		length = len(lastTimes)
-	}
+	length := min(len(lastTimes), len(perCoreTimes))
 
 	usage := make([]uint8, length)
 	for i := 0; i < length; i++ {

agent/data_dir.go

@@ -6,17 +6,19 @@ import (
 	"os"
 	"path/filepath"
 	"runtime"
+
+	"github.com/henrygd/beszel/agent/utils"
 )
 
-// getDataDir returns the path to the data directory for the agent and an error
+// GetDataDir returns the path to the data directory for the agent and an error
 // if the directory is not valid. Attempts to find the optimal data directory if
 // no data directories are provided.
-func getDataDir(dataDirs ...string) (string, error) {
+func GetDataDir(dataDirs ...string) (string, error) {
 	if len(dataDirs) > 0 {
 		return testDataDirs(dataDirs)
 	}
 
-	dataDir, _ := GetEnv("DATA_DIR")
+	dataDir, _ := utils.GetEnv("DATA_DIR")
 	if dataDir != "" {
 		dataDirs = append(dataDirs, dataDir)
 	}

agent/data_dir_test.go

@@ -1,5 +1,4 @@
 //go:build testing
-// +build testing
 
 package agent
 
@@ -17,7 +16,7 @@ func TestGetDataDir(t *testing.T) {
 	// Test with explicit dataDir parameter
 	t.Run("explicit data dir", func(t *testing.T) {
 		tempDir := t.TempDir()
-		result, err := getDataDir(tempDir)
+		result, err := GetDataDir(tempDir)
 		require.NoError(t, err)
 		assert.Equal(t, tempDir, result)
 	})
@@ -26,7 +25,7 @@ func TestGetDataDir(t *testing.T) {
 	t.Run("explicit data dir - create new", func(t *testing.T) {
 		tempDir := t.TempDir()
 		newDir := filepath.Join(tempDir, "new-data-dir")
-		result, err := getDataDir(newDir)
+		result, err := GetDataDir(newDir)
 		require.NoError(t, err)
 		assert.Equal(t, newDir, result)
 
@@ -40,19 +39,9 @@ func TestGetDataDir(t *testing.T) {
 	t.Run("DATA_DIR environment variable", func(t *testing.T) {
 		tempDir := t.TempDir()
 
-		// Set environment variable
-		oldValue := os.Getenv("DATA_DIR")
-		defer func() {
-			if oldValue == "" {
-				os.Unsetenv("BESZEL_AGENT_DATA_DIR")
-			} else {
-				os.Setenv("BESZEL_AGENT_DATA_DIR", oldValue)
-			}
-		}()
+		t.Setenv("BESZEL_AGENT_DATA_DIR", tempDir)
 
-		os.Setenv("BESZEL_AGENT_DATA_DIR", tempDir)
-
-		result, err := getDataDir()
+		result, err := GetDataDir()
 		require.NoError(t, err)
 		assert.Equal(t, tempDir, result)
 	})
@@ -60,26 +49,15 @@ func TestGetDataDir(t *testing.T) {
 	// Test with invalid explicit dataDir
 	t.Run("invalid explicit data dir", func(t *testing.T) {
 		invalidPath := "/invalid/path/that/cannot/be/created"
-		_, err := getDataDir(invalidPath)
+		_, err := GetDataDir(invalidPath)
 		assert.Error(t, err)
 	})
 
 	// Test fallback behavior (empty dataDir, no env var)
 	t.Run("fallback to default directories", func(t *testing.T) {
-		// Clear DATA_DIR environment variable
-		oldValue := os.Getenv("DATA_DIR")
-		defer func() {
-			if oldValue == "" {
-				os.Unsetenv("DATA_DIR")
-			} else {
-				os.Setenv("DATA_DIR", oldValue)
-			}
-		}()
-		os.Unsetenv("DATA_DIR")
-
 		// This will try platform-specific defaults, which may or may not work
 		// We're mainly testing that it doesn't panic and returns some result
-		result, err := getDataDir()
+		result, err := GetDataDir()
 		// We don't assert success/failure here since it depends on system permissions
 		// Just verify we get a string result if no error
 		if err == nil {

agent/disk.go

@@ -1,6 +1,7 @@
 package agent
 
 import (
+	"context"
 	"log/slog"
 	"os"
 	"path/filepath"
@@ -8,11 +9,59 @@ import (
 	"strings"
 	"time"
 
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"
 
 	"github.com/shirou/gopsutil/v4/disk"
 )
 
+// fsRegistrationContext holds the shared lookup state needed to resolve a
+// filesystem into the tracked fsStats key and metadata.
+type fsRegistrationContext struct {
+	filesystem     string // value of optional FILESYSTEM env var
+	isWindows      bool
+	efPath         string // path to extra filesystems (default "/extra-filesystems")
+	diskIoCounters map[string]disk.IOCountersStat
+}
+
+// diskDiscovery groups the transient state for a single initializeDiskInfo run so
+// helper methods can share the same partitions, mount paths, and lookup functions
+type diskDiscovery struct {
+	agent          *Agent
+	rootMountPoint string
+	partitions     []disk.PartitionStat
+	usageFn        func(string) (*disk.UsageStat, error)
+	ctx            fsRegistrationContext
+}
+
+// prevDisk stores previous per-device disk counters for a given cache interval
+type prevDisk struct {
+	readBytes  uint64
+	writeBytes uint64
+	readTime   uint64 // cumulative ms spent on reads (from ReadTime)
+	writeTime  uint64 // cumulative ms spent on writes (from WriteTime)
+	ioTime     uint64 // cumulative ms spent doing I/O (from IoTime)
+	weightedIO uint64 // cumulative weighted ms (queue-depth × ms, from WeightedIO)
+	readCount  uint64 // cumulative read operation count
+	writeCount uint64 // cumulative write operation count
+	at         time.Time
+}
+
+// prevDiskFromCounter creates a prevDisk snapshot from a disk.IOCountersStat at time t.
+func prevDiskFromCounter(d disk.IOCountersStat, t time.Time) prevDisk {
+	return prevDisk{
+		readBytes:  d.ReadBytes,
+		writeBytes: d.WriteBytes,
+		readTime:   d.ReadTime,
+		writeTime:  d.WriteTime,
+		ioTime:     d.IoTime,
+		weightedIO: d.WeightedIO,
+		readCount:  d.ReadCount,
+		writeCount: d.WriteCount,
+		at:         t,
+	}
+}
+
 // parseFilesystemEntry parses a filesystem entry in the format "device__customname"
 // Returns the device/filesystem part and the custom name part
 func parseFilesystemEntry(entry string) (device, customName string) {
@@ -26,14 +75,236 @@ func parseFilesystemEntry(entry string) (device, customName string) {
 	return device, customName
 }
 
+// extraFilesystemPartitionInfo derives the I/O device and optional display name
+// for a mounted /extra-filesystems partition. Prefer the partition device reported
+// by the system and only use the folder name for custom naming metadata.
+func extraFilesystemPartitionInfo(p disk.PartitionStat) (device, customName string) {
+	device = strings.TrimSpace(p.Device)
+	folderDevice, customName := parseFilesystemEntry(filepath.Base(p.Mountpoint))
+	if device == "" {
+		device = folderDevice
+	}
+	return device, customName
+}
+
+func isDockerSpecialMountpoint(mountpoint string) bool {
+	switch mountpoint {
+	case "/etc/hosts", "/etc/resolv.conf", "/etc/hostname":
+		return true
+	}
+	return false
+}
+
+// registerFilesystemStats resolves the tracked key and stats payload for a
+// filesystem before it is inserted into fsStats.
+func registerFilesystemStats(existing map[string]*system.FsStats, device, mountpoint string, root bool, customName string, ctx fsRegistrationContext) (string, *system.FsStats, bool) {
+	key := device
+	if !ctx.isWindows {
+		key = filepath.Base(device)
+	}
+
+	if root {
+		// Try to map root device to a diskIoCounters entry. First checks for an
+		// exact key match, then uses findIoDevice for normalized / prefix-based
+		// matching (e.g. nda0p2 -> nda0), and finally falls back to FILESYSTEM.
+		if _, ioMatch := ctx.diskIoCounters[key]; !ioMatch {
+			if matchedKey, match := findIoDevice(key, ctx.diskIoCounters); match {
+				key = matchedKey
+			} else if ctx.filesystem != "" {
+				if matchedKey, match := findIoDevice(ctx.filesystem, ctx.diskIoCounters); match {
+					key = matchedKey
+				}
+			}
+			if _, ioMatch = ctx.diskIoCounters[key]; !ioMatch {
+				slog.Warn("Root I/O unmapped; set FILESYSTEM", "device", device, "mountpoint", mountpoint)
+			}
+		}
+	} else {
+		// Check if non-root has diskstats and prefer the folder device for
+		// /extra-filesystems mounts when the discovered partition device is a
+		// mapper path (e.g. luks UUID) that obscures the underlying block device.
+		if _, ioMatch := ctx.diskIoCounters[key]; !ioMatch {
+			if strings.HasPrefix(mountpoint, ctx.efPath) {
+				folderDevice, _ := parseFilesystemEntry(filepath.Base(mountpoint))
+				if folderDevice != "" {
+					if matchedKey, match := findIoDevice(folderDevice, ctx.diskIoCounters); match {
+						key = matchedKey
+					}
+				}
+			}
+			if _, ioMatch = ctx.diskIoCounters[key]; !ioMatch {
+				if matchedKey, match := findIoDevice(key, ctx.diskIoCounters); match {
+					key = matchedKey
+				}
+			}
+		}
+	}
+
+	if _, exists := existing[key]; exists {
+		return "", nil, false
+	}
+
+	fsStats := &system.FsStats{Root: root, Mountpoint: mountpoint}
+	if customName != "" {
+		fsStats.Name = customName
+	}
+	return key, fsStats, true
+}
+
+// addFsStat inserts a discovered filesystem if it resolves to a new tracking
+// key. The key selection itself lives in buildFsStatRegistration so that logic
+// can stay directly unit-tested.
+func (d *diskDiscovery) addFsStat(device, mountpoint string, root bool, customName string) {
+	key, fsStats, ok := registerFilesystemStats(d.agent.fsStats, device, mountpoint, root, customName, d.ctx)
+	if !ok {
+		return
+	}
+	d.agent.fsStats[key] = fsStats
+	name := key
+	if customName != "" {
+		name = customName
+	}
+	slog.Info("Detected disk", "name", name, "device", device, "mount", mountpoint, "io", key, "root", root)
+}
+
+// addConfiguredRootFs resolves FILESYSTEM against partitions first, then falls
+// back to direct diskstats matching for setups like ZFS where partitions do not
+// expose the physical device name.
+func (d *diskDiscovery) addConfiguredRootFs() bool {
+	if d.ctx.filesystem == "" {
+		return false
+	}
+
+	for _, p := range d.partitions {
+		if filesystemMatchesPartitionSetting(d.ctx.filesystem, p) {
+			d.addFsStat(p.Device, p.Mountpoint, true, "")
+			return true
+		}
+	}
+
+	// FILESYSTEM may name a physical disk absent from partitions (e.g. ZFS lists
+	// dataset paths like zroot/ROOT/default, not block devices).
+	if ioKey, match := findIoDevice(d.ctx.filesystem, d.ctx.diskIoCounters); match {
+		d.agent.fsStats[ioKey] = &system.FsStats{Root: true, Mountpoint: d.rootMountPoint}
+		return true
+	}
+
+	slog.Warn("Partition details not found", "filesystem", d.ctx.filesystem)
+	return false
+}
+
+func isRootFallbackPartition(p disk.PartitionStat, rootMountPoint string) bool {
+	return p.Mountpoint == rootMountPoint ||
+		(isDockerSpecialMountpoint(p.Mountpoint) && strings.HasPrefix(p.Device, "/dev"))
+}
+
+// addPartitionRootFs handles the non-configured root fallback path when a
+// partition looks like the active root mount but still needs translating to an
+// I/O device key.
+func (d *diskDiscovery) addPartitionRootFs(device, mountpoint string) bool {
+	fs, match := findIoDevice(filepath.Base(device), d.ctx.diskIoCounters)
+	if !match {
+		return false
+	}
+	// The resolved I/O device is already known here, so use it directly to avoid
+	// a second fallback search inside buildFsStatRegistration.
+	d.addFsStat(fs, mountpoint, true, "")
+	return true
+}
+
+// addLastResortRootFs is only used when neither FILESYSTEM nor partition-based
+// heuristics can identify root, so it picks the busiest I/O device as a final
+// fallback and preserves the root mountpoint for usage collection.
+func (d *diskDiscovery) addLastResortRootFs() {
+	rootKey := mostActiveIoDevice(d.ctx.diskIoCounters)
+	if rootKey != "" {
+		slog.Warn("Using most active device for root I/O; set FILESYSTEM to override", "device", rootKey)
+	} else {
+		rootKey = filepath.Base(d.rootMountPoint)
+		if _, exists := d.agent.fsStats[rootKey]; exists {
+			rootKey = "root"
+		}
+		slog.Warn("Root I/O device not detected; set FILESYSTEM to override")
+	}
+	d.agent.fsStats[rootKey] = &system.FsStats{Root: true, Mountpoint: d.rootMountPoint}
+}
+
+// findPartitionByFilesystemSetting matches an EXTRA_FILESYSTEMS entry against a
+// discovered partition either by mountpoint or by device suffix.
+func findPartitionByFilesystemSetting(filesystem string, partitions []disk.PartitionStat) (disk.PartitionStat, bool) {
+	for _, p := range partitions {
+		if strings.HasSuffix(p.Device, filesystem) || p.Mountpoint == filesystem {
+			return p, true
+		}
+	}
+	return disk.PartitionStat{}, false
+}
+
+// addConfiguredExtraFsEntry resolves one EXTRA_FILESYSTEMS entry, preferring a
+// discovered partition and falling back to any path that disk.Usage accepts.
+func (d *diskDiscovery) addConfiguredExtraFsEntry(filesystem, customName string) {
+	if p, found := findPartitionByFilesystemSetting(filesystem, d.partitions); found {
+		d.addFsStat(p.Device, p.Mountpoint, false, customName)
+		return
+	}
+
+	if _, err := d.usageFn(filesystem); err == nil {
+		d.addFsStat(filepath.Base(filesystem), filesystem, false, customName)
+		return
+	} else {
+		slog.Error("Invalid filesystem", "name", filesystem, "err", err)
+	}
+}
+
+// addConfiguredExtraFilesystems parses and registers the comma-separated
+// EXTRA_FILESYSTEMS env var entries.
+func (d *diskDiscovery) addConfiguredExtraFilesystems(extraFilesystems string) {
+	for fsEntry := range strings.SplitSeq(extraFilesystems, ",") {
+		filesystem, customName := parseFilesystemEntry(fsEntry)
+		d.addConfiguredExtraFsEntry(filesystem, customName)
+	}
+}
+
+// addPartitionExtraFs registers partitions mounted under /extra-filesystems so
+// their display names can come from the folder name while their I/O keys still
+// prefer the underlying partition device. Only direct children are matched to
+// avoid registering nested virtual mounts (e.g. /proc, /sys) that are returned by
+// disk.Partitions(true) when the host root is bind-mounted in /extra-filesystems.
+func (d *diskDiscovery) addPartitionExtraFs(p disk.PartitionStat) {
+	if filepath.Dir(p.Mountpoint) != d.ctx.efPath {
+		return
+	}
+	device, customName := extraFilesystemPartitionInfo(p)
+	d.addFsStat(device, p.Mountpoint, false, customName)
+}
+
+// addExtraFilesystemFolders handles bare directories under /extra-filesystems
+// that may not appear in partition discovery, while skipping mountpoints that
+// were already registered from higher-fidelity sources.
+func (d *diskDiscovery) addExtraFilesystemFolders(folderNames []string) {
+	existingMountpoints := make(map[string]bool, len(d.agent.fsStats))
+	for _, stats := range d.agent.fsStats {
+		existingMountpoints[stats.Mountpoint] = true
+	}
+
+	for _, folderName := range folderNames {
+		mountpoint := filepath.Join(d.ctx.efPath, folderName)
+		slog.Debug("/extra-filesystems", "mountpoint", mountpoint)
+		if existingMountpoints[mountpoint] {
+			continue
+		}
+		device, customName := parseFilesystemEntry(folderName)
+		d.addFsStat(device, mountpoint, false, customName)
+	}
+}
+
 // Sets up the filesystems to monitor for disk usage and I/O.
 func (a *Agent) initializeDiskInfo() {
-	filesystem, _ := GetEnv("FILESYSTEM")
-	efPath := "/extra-filesystems"
+	filesystem, _ := utils.GetEnv("FILESYSTEM")
 	hasRoot := false
 	isWindows := runtime.GOOS == "windows"
 
-	partitions, err := disk.Partitions(false)
+	partitions, err := disk.PartitionsWithContext(context.Background(), true)
 	if err != nil {
 		slog.Error("Error getting disk partitions", "err", err)
 	}
@@ -46,167 +317,223 @@ func (a *Agent) initializeDiskInfo() {
 		}
 	}
 
-	// ioContext := context.WithValue(a.sensorsContext,
-	// 	common.EnvKey, common.EnvMap{common.HostProcEnvKey: "/tmp/testproc"},
-	// )
-	// diskIoCounters, err := disk.IOCountersWithContext(ioContext)
-
 	diskIoCounters, err := disk.IOCounters()
 	if err != nil {
 		slog.Error("Error getting diskstats", "err", err)
 	}
 	slog.Debug("Disk I/O", "diskstats", diskIoCounters)
-
-	// Helper function to add a filesystem to fsStats if it doesn't exist
-	addFsStat := func(device, mountpoint string, root bool, customName ...string) {
-		var key string
-		if isWindows {
-			key = device
-		} else {
-			key = filepath.Base(device)
-		}
-		var ioMatch bool
-		if _, exists := a.fsStats[key]; !exists {
-			if root {
-				slog.Info("Detected root device", "name", key)
-				// Check if root device is in /proc/diskstats, use fallback if not
-				if _, ioMatch = diskIoCounters[key]; !ioMatch {
-					key, ioMatch = findIoDevice(filesystem, diskIoCounters, a.fsStats)
-					if !ioMatch {
-						slog.Info("Using I/O fallback", "device", device, "mountpoint", mountpoint, "fallback", key)
-					}
-				}
-			} else {
-				// Check if non-root has diskstats and fall back to folder name if not
-				// Scenario: device is encrypted and named luks-2bcb02be-999d-4417-8d18-5c61e660fb6e - not in /proc/diskstats.
-				// However, the device can be specified by mounting folder from luks device at /extra-filesystems/sda1
-				if _, ioMatch = diskIoCounters[key]; !ioMatch {
-					efBase := filepath.Base(mountpoint)
-					if _, ioMatch = diskIoCounters[efBase]; ioMatch {
-						key = efBase
-					}
-				}
-			}
-			fsStats := &system.FsStats{Root: root, Mountpoint: mountpoint}
-			if len(customName) > 0 && customName[0] != "" {
-				fsStats.Name = customName[0]
-			}
-			a.fsStats[key] = fsStats
-		}
+	ctx := fsRegistrationContext{
+		filesystem:     filesystem,
+		isWindows:      isWindows,
+		diskIoCounters: diskIoCounters,
+		efPath:         "/extra-filesystems",
 	}
 
 	// Get the appropriate root mount point for this system
-	rootMountPoint := a.getRootMountPoint()
-
-	// Use FILESYSTEM env var to find root filesystem
-	if filesystem != "" {
-		for _, p := range partitions {
-			if strings.HasSuffix(p.Device, filesystem) || p.Mountpoint == filesystem {
-				addFsStat(p.Device, p.Mountpoint, true)
-				hasRoot = true
-				break
-			}
-		}
-		if !hasRoot {
-			slog.Warn("Partition details not found", "filesystem", filesystem)
-		}
+	discovery := diskDiscovery{
+		agent:          a,
+		rootMountPoint: a.getRootMountPoint(),
+		partitions:     partitions,
+		usageFn:        disk.Usage,
+		ctx:            ctx,
 	}
 
-	// Add EXTRA_FILESYSTEMS env var values to fsStats
-	if extraFilesystems, exists := GetEnv("EXTRA_FILESYSTEMS"); exists {
-		for _, fsEntry := range strings.Split(extraFilesystems, ",") {
-			// Parse custom name from format: device__customname
-			fs, customName := parseFilesystemEntry(fsEntry)
+	hasRoot = discovery.addConfiguredRootFs()
 
-			found := false
-			for _, p := range partitions {
-				if strings.HasSuffix(p.Device, fs) || p.Mountpoint == fs {
-					addFsStat(p.Device, p.Mountpoint, false, customName)
-					found = true
-					break
-				}
-			}
-			// if not in partitions, test if we can get disk usage
-			if !found {
-				if _, err := disk.Usage(fs); err == nil {
-					addFsStat(filepath.Base(fs), fs, false, customName)
-				} else {
-					slog.Error("Invalid filesystem", "name", fs, "err", err)
-				}
-			}
-		}
+	// Add EXTRA_FILESYSTEMS env var values to fsStats
+	if extraFilesystems, exists := utils.GetEnv("EXTRA_FILESYSTEMS"); exists {
+		discovery.addConfiguredExtraFilesystems(extraFilesystems)
 	}
 
 	// Process partitions for various mount points
 	for _, p := range partitions {
-		// fmt.Println(p.Device, p.Mountpoint)
-		// Binary root fallback or docker root fallback
-		if !hasRoot && (p.Mountpoint == rootMountPoint || (p.Mountpoint == "/etc/hosts" && strings.HasPrefix(p.Device, "/dev"))) {
-			fs, match := findIoDevice(filepath.Base(p.Device), diskIoCounters, a.fsStats)
-			if match {
-				addFsStat(fs, p.Mountpoint, true)
-				hasRoot = true
-			}
-		}
-
-		// Check if device is in /extra-filesystems
-		if strings.HasPrefix(p.Mountpoint, efPath) {
-			device, customName := parseFilesystemEntry(p.Mountpoint)
-			addFsStat(device, p.Mountpoint, false, customName)
+		if !hasRoot && isRootFallbackPartition(p, discovery.rootMountPoint) {
+			hasRoot = discovery.addPartitionRootFs(p.Device, p.Mountpoint)
 		}
+		discovery.addPartitionExtraFs(p)
 	}
 
 	// Check all folders in /extra-filesystems and add them if not already present
-	if folders, err := os.ReadDir(efPath); err == nil {
-		existingMountpoints := make(map[string]bool)
-		for _, stats := range a.fsStats {
-			existingMountpoints[stats.Mountpoint] = true
-		}
+	if folders, err := os.ReadDir(discovery.ctx.efPath); err == nil {
+		folderNames := make([]string, 0, len(folders))
 		for _, folder := range folders {
 			if folder.IsDir() {
-				mountpoint := filepath.Join(efPath, folder.Name())
-				slog.Debug("/extra-filesystems", "mountpoint", mountpoint)
-				if !existingMountpoints[mountpoint] {
-					device, customName := parseFilesystemEntry(folder.Name())
-					addFsStat(device, mountpoint, false, customName)
-				}
+				folderNames = append(folderNames, folder.Name())
 			}
 		}
+		discovery.addExtraFilesystemFolders(folderNames)
 	}
 
-	// If no root filesystem set, use fallback
+	// If no root filesystem set, try the most active I/O device as a last
+	// resort (e.g. ZFS where dataset names are unrelated to disk names).
 	if !hasRoot {
-		rootDevice, _ := findIoDevice(filepath.Base(filesystem), diskIoCounters, a.fsStats)
-		slog.Info("Root disk", "mountpoint", rootMountPoint, "io", rootDevice)
-		a.fsStats[rootDevice] = &system.FsStats{Root: true, Mountpoint: rootMountPoint}
+		discovery.addLastResortRootFs()
 	}
 
+	a.pruneDuplicateRootExtraFilesystems()
 	a.initializeDiskIoStats(diskIoCounters)
 }
 
-// Returns matching device from /proc/diskstats,
-// or the device with the most reads if no match is found.
-// bool is true if a match was found.
-func findIoDevice(filesystem string, diskIoCounters map[string]disk.IOCountersStat, fsStats map[string]*system.FsStats) (string, bool) {
-	var maxReadBytes uint64
-	maxReadDevice := "/"
-	for _, d := range diskIoCounters {
-		if d.Name == filesystem || (d.Label != "" && d.Label == filesystem) {
-			return d.Name, true
-		}
-		if d.ReadBytes > maxReadBytes {
-			// don't use if device already exists in fsStats
-			if _, exists := fsStats[d.Name]; !exists {
-				maxReadBytes = d.ReadBytes
-				maxReadDevice = d.Name
-			}
+// Removes extra filesystems that mirror root usage (https://github.com/henrygd/beszel/issues/1428).
+func (a *Agent) pruneDuplicateRootExtraFilesystems() {
+	var rootMountpoint string
+	for _, stats := range a.fsStats {
+		if stats != nil && stats.Root {
+			rootMountpoint = stats.Mountpoint
+			break
 		}
 	}
-	return maxReadDevice, false
+	if rootMountpoint == "" {
+		return
+	}
+	rootUsage, err := disk.Usage(rootMountpoint)
+	if err != nil {
+		return
+	}
+	for name, stats := range a.fsStats {
+		if stats == nil || stats.Root {
+			continue
+		}
+		extraUsage, err := disk.Usage(stats.Mountpoint)
+		if err != nil {
+			continue
+		}
+		if hasSameDiskUsage(rootUsage, extraUsage) {
+			slog.Info("Ignoring duplicate FS", "name", name, "mount", stats.Mountpoint)
+			delete(a.fsStats, name)
+		}
+	}
+}
+
+// hasSameDiskUsage compares root/extra usage with a small byte tolerance.
+func hasSameDiskUsage(a, b *disk.UsageStat) bool {
+	if a == nil || b == nil || a.Total == 0 || b.Total == 0 {
+		return false
+	}
+	// Allow minor drift between sequential disk usage calls.
+	const toleranceBytes uint64 = 16 * 1024 * 1024
+	return withinUsageTolerance(a.Total, b.Total, toleranceBytes) &&
+		withinUsageTolerance(a.Used, b.Used, toleranceBytes)
+}
+
+// withinUsageTolerance reports whether two byte values differ by at most tolerance.
+func withinUsageTolerance(a, b, tolerance uint64) bool {
+	if a >= b {
+		return a-b <= tolerance
+	}
+	return b-a <= tolerance
+}
+
+type ioMatchCandidate struct {
+	name  string
+	bytes uint64
+	ops   uint64
+}
+
+// findIoDevice prefers exact device/label matches, then falls back to a
+// prefix-related candidate with the highest recent activity.
+func findIoDevice(filesystem string, diskIoCounters map[string]disk.IOCountersStat) (string, bool) {
+	filesystem = normalizeDeviceName(filesystem)
+	if filesystem == "" {
+		return "", false
+	}
+
+	candidates := []ioMatchCandidate{}
+
+	for _, d := range diskIoCounters {
+		if normalizeDeviceName(d.Name) == filesystem || (d.Label != "" && normalizeDeviceName(d.Label) == filesystem) {
+			return d.Name, true
+		}
+		if prefixRelated(normalizeDeviceName(d.Name), filesystem) ||
+			(d.Label != "" && prefixRelated(normalizeDeviceName(d.Label), filesystem)) {
+			candidates = append(candidates, ioMatchCandidate{
+				name:  d.Name,
+				bytes: d.ReadBytes + d.WriteBytes,
+				ops:   d.ReadCount + d.WriteCount,
+			})
+		}
+	}
+
+	if len(candidates) == 0 {
+		return "", false
+	}
+
+	best := candidates[0]
+	for _, c := range candidates[1:] {
+		if c.bytes > best.bytes ||
+			(c.bytes == best.bytes && c.ops > best.ops) ||
+			(c.bytes == best.bytes && c.ops == best.ops && c.name < best.name) {
+			best = c
+		}
+	}
+
+	slog.Info("Using disk I/O fallback", "requested", filesystem, "selected", best.name)
+	return best.name, true
+}
+
+// mostActiveIoDevice returns the device with the highest I/O activity,
+// or "" if diskIoCounters is empty.
+func mostActiveIoDevice(diskIoCounters map[string]disk.IOCountersStat) string {
+	var best ioMatchCandidate
+	for _, d := range diskIoCounters {
+		c := ioMatchCandidate{
+			name:  d.Name,
+			bytes: d.ReadBytes + d.WriteBytes,
+			ops:   d.ReadCount + d.WriteCount,
+		}
+		if best.name == "" || c.bytes > best.bytes ||
+			(c.bytes == best.bytes && c.ops > best.ops) ||
+			(c.bytes == best.bytes && c.ops == best.ops && c.name < best.name) {
+			best = c
+		}
+	}
+	return best.name
+}
+
+// prefixRelated reports whether either identifier is a prefix of the other.
+func prefixRelated(a, b string) bool {
+	if a == "" || b == "" || a == b {
+		return false
+	}
+	return strings.HasPrefix(a, b) || strings.HasPrefix(b, a)
+}
+
+// filesystemMatchesPartitionSetting checks whether a FILESYSTEM env var value
+// matches a partition by mountpoint, exact device name, or prefix relationship
+// (e.g. FILESYSTEM=ada0 matches partition /dev/ada0p2).
+func filesystemMatchesPartitionSetting(filesystem string, p disk.PartitionStat) bool {
+	filesystem = strings.TrimSpace(filesystem)
+	if filesystem == "" {
+		return false
+	}
+	if p.Mountpoint == filesystem {
+		return true
+	}
+
+	fsName := normalizeDeviceName(filesystem)
+	partName := normalizeDeviceName(p.Device)
+	if fsName == "" || partName == "" {
+		return false
+	}
+	if fsName == partName {
+		return true
+	}
+	return prefixRelated(partName, fsName)
+}
+
+// normalizeDeviceName canonicalizes device strings for comparisons.
+func normalizeDeviceName(value string) string {
+	name := filepath.Base(strings.TrimSpace(value))
+	if name == "." {
+		return ""
+	}
+	return name
 }
 
 // Sets start values for disk I/O stats.
 func (a *Agent) initializeDiskIoStats(diskIoCounters map[string]disk.IOCountersStat) {
+	a.fsNames = a.fsNames[:0]
+	now := time.Now()
 	for device, stats := range a.fsStats {
 		// skip if not in diskIoCounters
 		d, exists := diskIoCounters[device]
@@ -215,7 +542,7 @@ func (a *Agent) initializeDiskIoStats(diskIoCounters map[string]disk.IOCountersS
 			continue
 		}
 		// populate initial values
-		stats.Time = time.Now()
+		stats.Time = now
 		stats.TotalRead = d.ReadBytes
 		stats.TotalWrite = d.WriteBytes
 		// add to list of valid io device names
@@ -239,12 +566,12 @@ func (a *Agent) updateDiskUsage(systemStats *system.Stats) {
 			continue
 		}
 		if d, err := disk.Usage(stats.Mountpoint); err == nil {
-			stats.DiskTotal = bytesToGigabytes(d.Total)
-			stats.DiskUsed = bytesToGigabytes(d.Used)
+			stats.DiskTotal = utils.BytesToGigabytes(d.Total)
+			stats.DiskUsed = utils.BytesToGigabytes(d.Used)
 			if stats.Root {
-				systemStats.DiskTotal = bytesToGigabytes(d.Total)
-				systemStats.DiskUsed = bytesToGigabytes(d.Used)
-				systemStats.DiskPct = twoDecimals(d.UsedPercent)
+				systemStats.DiskTotal = utils.BytesToGigabytes(d.Total)
+				systemStats.DiskUsed = utils.BytesToGigabytes(d.Used)
+				systemStats.DiskPct = utils.TwoDecimals(d.UsedPercent)
 			}
 		} else {
 			// reset stats if error (likely unmounted)
@@ -282,36 +609,72 @@ func (a *Agent) updateDiskIo(cacheTimeMs uint16, systemStats *system.Stats) {
 			prev, hasPrev := a.diskPrev[cacheTimeMs][name]
 			if !hasPrev {
 				// Seed from agent-level fsStats if present, else seed from current
-				prev = prevDisk{readBytes: stats.TotalRead, writeBytes: stats.TotalWrite, at: stats.Time}
+				prev = prevDisk{
+					readBytes:  stats.TotalRead,
+					writeBytes: stats.TotalWrite,
+					readTime:   d.ReadTime,
+					writeTime:  d.WriteTime,
+					ioTime:     d.IoTime,
+					weightedIO: d.WeightedIO,
+					readCount:  d.ReadCount,
+					writeCount: d.WriteCount,
+					at:         stats.Time,
+				}
 				if prev.at.IsZero() {
-					prev = prevDisk{readBytes: d.ReadBytes, writeBytes: d.WriteBytes, at: now}
+					prev = prevDiskFromCounter(d, now)
 				}
 			}
 
 			msElapsed := uint64(now.Sub(prev.at).Milliseconds())
+
+			// Update per-interval snapshot
+			a.diskPrev[cacheTimeMs][name] = prevDiskFromCounter(d, now)
+
+			// Avoid division by zero or clock issues
 			if msElapsed < 100 {
-				// Avoid division by zero or clock issues; update snapshot and continue
-				a.diskPrev[cacheTimeMs][name] = prevDisk{readBytes: d.ReadBytes, writeBytes: d.WriteBytes, at: now}
 				continue
 			}
 
 			diskIORead := (d.ReadBytes - prev.readBytes) * 1000 / msElapsed
 			diskIOWrite := (d.WriteBytes - prev.writeBytes) * 1000 / msElapsed
-			readMbPerSecond := bytesToMegabytes(float64(diskIORead))
-			writeMbPerSecond := bytesToMegabytes(float64(diskIOWrite))
+			readMbPerSecond := utils.BytesToMegabytes(float64(diskIORead))
+			writeMbPerSecond := utils.BytesToMegabytes(float64(diskIOWrite))
 
 			// validate values
 			if readMbPerSecond > 50_000 || writeMbPerSecond > 50_000 {
 				slog.Warn("Invalid disk I/O. Resetting.", "name", d.Name, "read", readMbPerSecond, "write", writeMbPerSecond)
-				// Reset interval snapshot and seed from current
-				a.diskPrev[cacheTimeMs][name] = prevDisk{readBytes: d.ReadBytes, writeBytes: d.WriteBytes, at: now}
 				// also refresh agent baseline to avoid future negatives
 				a.initializeDiskIoStats(ioCounters)
 				continue
 			}
 
-			// Update per-interval snapshot
-			a.diskPrev[cacheTimeMs][name] = prevDisk{readBytes: d.ReadBytes, writeBytes: d.WriteBytes, at: now}
+			// These properties are calculated differently on different platforms,
+			// but generally represent cumulative time spent doing reads/writes on the device.
+			// This can surpass 100% if there are multiple concurrent I/O operations.
+			// Linux kernel docs:
+			// This is the total number of milliseconds spent by all reads (as
+			// measured from __make_request() to end_that_request_last()).
+			// https://www.kernel.org/doc/Documentation/iostats.txt (fields 4, 8)
+			diskReadTime := utils.TwoDecimals(float64(d.ReadTime-prev.readTime) / float64(msElapsed) * 100)
+			diskWriteTime := utils.TwoDecimals(float64(d.WriteTime-prev.writeTime) / float64(msElapsed) * 100)
+
+			// I/O utilization %: fraction of wall time the device had any I/O in progress (0-100).
+			diskIoUtilPct := utils.TwoDecimals(float64(d.IoTime-prev.ioTime) / float64(msElapsed) * 100)
+
+			// Weighted I/O: queue-depth weighted I/O time, normalized to interval (can exceed 100%).
+			// Linux kernel field 11: incremented by iops_in_progress × ms_since_last_update.
+			// Used to display queue depth. Multipled by 100 to increase accuracy of digit truncation (divided by 100 in UI).
+			diskWeightedIO := utils.TwoDecimals(float64(d.WeightedIO-prev.weightedIO) / float64(msElapsed) * 100)
+
+			// r_await / w_await: average time per read/write operation in milliseconds.
+			// Equivalent to r_await and w_await in iostat.
+			var rAwait, wAwait float64
+			if deltaReadCount := d.ReadCount - prev.readCount; deltaReadCount > 0 {
+				rAwait = utils.TwoDecimals(float64(d.ReadTime-prev.readTime) / float64(deltaReadCount))
+			}
+			if deltaWriteCount := d.WriteCount - prev.writeCount; deltaWriteCount > 0 {
+				wAwait = utils.TwoDecimals(float64(d.WriteTime-prev.writeTime) / float64(deltaWriteCount))
+			}
 
 			// Update global fsStats baseline for cross-interval correctness
 			stats.Time = now
@@ -321,20 +684,40 @@ func (a *Agent) updateDiskIo(cacheTimeMs uint16, systemStats *system.Stats) {
 			stats.DiskWritePs = writeMbPerSecond
 			stats.DiskReadBytes = diskIORead
 			stats.DiskWriteBytes = diskIOWrite
+			stats.DiskIoStats[0] = diskReadTime
+			stats.DiskIoStats[1] = diskWriteTime
+			stats.DiskIoStats[2] = diskIoUtilPct
+			stats.DiskIoStats[3] = rAwait
+			stats.DiskIoStats[4] = wAwait
+			stats.DiskIoStats[5] = diskWeightedIO
 
 			if stats.Root {
 				systemStats.DiskReadPs = stats.DiskReadPs
 				systemStats.DiskWritePs = stats.DiskWritePs
 				systemStats.DiskIO[0] = diskIORead
 				systemStats.DiskIO[1] = diskIOWrite
+				systemStats.DiskIoStats[0] = diskReadTime
+				systemStats.DiskIoStats[1] = diskWriteTime
+				systemStats.DiskIoStats[2] = diskIoUtilPct
+				systemStats.DiskIoStats[3] = rAwait
+				systemStats.DiskIoStats[4] = wAwait
+				systemStats.DiskIoStats[5] = diskWeightedIO
 			}
 		}
 	}
 }
 
-// getRootMountPoint returns the appropriate root mount point for the system
+// getRootMountPoint returns the appropriate root mount point for the system.
+// On Windows it returns the system drive (e.g. "C:").
 // For immutable systems like Fedora Silverblue, it returns /sysroot instead of /
 func (a *Agent) getRootMountPoint() string {
+	if runtime.GOOS == "windows" {
+		if sd := os.Getenv("SystemDrive"); sd != "" {
+			return sd
+		}
+		return "C:"
+	}
+
 	// 1. Check if /etc/os-release contains indicators of an immutable system
 	if osReleaseContent, err := os.ReadFile("/etc/os-release"); err == nil {
 		content := string(osReleaseContent)

agent/disk_test.go

@@ -1,5 +1,4 @@
 //go:build testing
-// +build testing
 
 package agent
 
@@ -94,19 +93,683 @@ func TestParseFilesystemEntry(t *testing.T) {
 	}
 }
 
-func TestInitializeDiskInfoWithCustomNames(t *testing.T) {
-	// Set up environment variables
-	oldEnv := os.Getenv("EXTRA_FILESYSTEMS")
-	defer func() {
-		if oldEnv != "" {
-			os.Setenv("EXTRA_FILESYSTEMS", oldEnv)
-		} else {
-			os.Unsetenv("EXTRA_FILESYSTEMS")
-		}
-	}()
+func TestExtraFilesystemPartitionInfo(t *testing.T) {
+	t.Run("uses partition device for label-only mountpoint", func(t *testing.T) {
+		device, customName := extraFilesystemPartitionInfo(disk.PartitionStat{
+			Device:     "/dev/sdc",
+			Mountpoint: "/extra-filesystems/Share",
+		})
 
+		assert.Equal(t, "/dev/sdc", device)
+		assert.Equal(t, "", customName)
+	})
+
+	t.Run("uses custom name from mountpoint suffix", func(t *testing.T) {
+		device, customName := extraFilesystemPartitionInfo(disk.PartitionStat{
+			Device:     "/dev/sdc",
+			Mountpoint: "/extra-filesystems/sdc__Share",
+		})
+
+		assert.Equal(t, "/dev/sdc", device)
+		assert.Equal(t, "Share", customName)
+	})
+
+	t.Run("falls back to folder device when partition device is unavailable", func(t *testing.T) {
+		device, customName := extraFilesystemPartitionInfo(disk.PartitionStat{
+			Mountpoint: "/extra-filesystems/sdc__Share",
+		})
+
+		assert.Equal(t, "sdc", device)
+		assert.Equal(t, "Share", customName)
+	})
+
+	t.Run("supports custom name without folder device prefix", func(t *testing.T) {
+		device, customName := extraFilesystemPartitionInfo(disk.PartitionStat{
+			Device:     "/dev/sdc",
+			Mountpoint: "/extra-filesystems/__Share",
+		})
+
+		assert.Equal(t, "/dev/sdc", device)
+		assert.Equal(t, "Share", customName)
+	})
+}
+
+func TestBuildFsStatRegistration(t *testing.T) {
+	t.Run("uses basename for non-windows exact io match", func(t *testing.T) {
+		key, stats, ok := registerFilesystemStats(
+			map[string]*system.FsStats{},
+			"/dev/sda1",
+			"/mnt/data",
+			false,
+			"archive",
+			fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"sda1": {Name: "sda1"},
+				},
+			},
+		)
+
+		assert.True(t, ok)
+		assert.Equal(t, "sda1", key)
+		assert.Equal(t, "/mnt/data", stats.Mountpoint)
+		assert.Equal(t, "archive", stats.Name)
+		assert.False(t, stats.Root)
+	})
+
+	t.Run("maps root partition to io device by prefix", func(t *testing.T) {
+		key, stats, ok := registerFilesystemStats(
+			map[string]*system.FsStats{},
+			"/dev/ada0p2",
+			"/",
+			true,
+			"",
+			fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"ada0": {Name: "ada0", ReadBytes: 1000, WriteBytes: 1000},
+				},
+			},
+		)
+
+		assert.True(t, ok)
+		assert.Equal(t, "ada0", key)
+		assert.True(t, stats.Root)
+		assert.Equal(t, "/", stats.Mountpoint)
+	})
+
+	t.Run("uses filesystem setting as root fallback", func(t *testing.T) {
+		key, _, ok := registerFilesystemStats(
+			map[string]*system.FsStats{},
+			"overlay",
+			"/",
+			true,
+			"",
+			fsRegistrationContext{
+				filesystem: "nvme0n1p2",
+				isWindows:  false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"nvme0n1": {Name: "nvme0n1", ReadBytes: 1000, WriteBytes: 1000},
+				},
+			},
+		)
+
+		assert.True(t, ok)
+		assert.Equal(t, "nvme0n1", key)
+	})
+
+	t.Run("prefers parsed extra-filesystems device over mapper device", func(t *testing.T) {
+		key, stats, ok := registerFilesystemStats(
+			map[string]*system.FsStats{},
+			"/dev/mapper/luks-2bcb02be-999d-4417-8d18-5c61e660fb6e",
+			"/extra-filesystems/nvme0n1p2__Archive",
+			false,
+			"Archive",
+			fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"dm-1":      {Name: "dm-1", Label: "luks-2bcb02be-999d-4417-8d18-5c61e660fb6e"},
+					"nvme0n1p2": {Name: "nvme0n1p2"},
+				},
+			},
+		)
+
+		assert.True(t, ok)
+		assert.Equal(t, "nvme0n1p2", key)
+		assert.Equal(t, "Archive", stats.Name)
+	})
+
+	t.Run("falls back to mapper io device when folder device cannot be resolved", func(t *testing.T) {
+		key, stats, ok := registerFilesystemStats(
+			map[string]*system.FsStats{},
+			"/dev/mapper/luks-2bcb02be-999d-4417-8d18-5c61e660fb6e",
+			"/extra-filesystems/Archive",
+			false,
+			"Archive",
+			fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"dm-1": {Name: "dm-1", Label: "luks-2bcb02be-999d-4417-8d18-5c61e660fb6e"},
+				},
+			},
+		)
+
+		assert.True(t, ok)
+		assert.Equal(t, "dm-1", key)
+		assert.Equal(t, "Archive", stats.Name)
+	})
+
+	t.Run("uses full device name on windows", func(t *testing.T) {
+		key, _, ok := registerFilesystemStats(
+			map[string]*system.FsStats{},
+			`C:`,
+			`C:\\`,
+			false,
+			"",
+			fsRegistrationContext{
+				isWindows: true,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					`C:`: {Name: `C:`},
+				},
+			},
+		)
+
+		assert.True(t, ok)
+		assert.Equal(t, `C:`, key)
+	})
+
+	t.Run("skips existing key", func(t *testing.T) {
+		key, stats, ok := registerFilesystemStats(
+			map[string]*system.FsStats{"sda1": {Mountpoint: "/existing"}},
+			"/dev/sda1",
+			"/mnt/data",
+			false,
+			"",
+			fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"sda1": {Name: "sda1"},
+				},
+			},
+		)
+
+		assert.False(t, ok)
+		assert.Empty(t, key)
+		assert.Nil(t, stats)
+	})
+}
+
+func TestAddConfiguredRootFs(t *testing.T) {
+	t.Run("adds root from matching partition", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent:          agent,
+			rootMountPoint: "/",
+			partitions:     []disk.PartitionStat{{Device: "/dev/ada0p2", Mountpoint: "/"}},
+			ctx: fsRegistrationContext{
+				filesystem: "/dev/ada0p2",
+				isWindows:  false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"ada0": {Name: "ada0", ReadBytes: 1000, WriteBytes: 1000},
+				},
+			},
+		}
+
+		ok := discovery.addConfiguredRootFs()
+
+		assert.True(t, ok)
+		stats, exists := agent.fsStats["ada0"]
+		assert.True(t, exists)
+		assert.True(t, stats.Root)
+		assert.Equal(t, "/", stats.Mountpoint)
+	})
+
+	t.Run("adds root from io device when partition is missing", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent:          agent,
+			rootMountPoint: "/sysroot",
+			ctx: fsRegistrationContext{
+				filesystem: "zroot",
+				isWindows:  false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"nda0": {Name: "nda0", Label: "zroot", ReadBytes: 1000, WriteBytes: 1000},
+				},
+			},
+		}
+
+		ok := discovery.addConfiguredRootFs()
+
+		assert.True(t, ok)
+		stats, exists := agent.fsStats["nda0"]
+		assert.True(t, exists)
+		assert.True(t, stats.Root)
+		assert.Equal(t, "/sysroot", stats.Mountpoint)
+	})
+
+	t.Run("returns false when filesystem cannot be resolved", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent:          agent,
+			rootMountPoint: "/",
+			ctx: fsRegistrationContext{
+				filesystem:     "missing-disk",
+				isWindows:      false,
+				diskIoCounters: map[string]disk.IOCountersStat{},
+			},
+		}
+
+		ok := discovery.addConfiguredRootFs()
+
+		assert.False(t, ok)
+		assert.Empty(t, agent.fsStats)
+	})
+}
+
+func TestAddPartitionRootFs(t *testing.T) {
+	t.Run("adds root from fallback partition candidate", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent: agent,
+			ctx: fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"nvme0n1": {Name: "nvme0n1", ReadBytes: 1000, WriteBytes: 1000},
+				},
+			},
+		}
+
+		ok := discovery.addPartitionRootFs("/dev/nvme0n1p2", "/")
+
+		assert.True(t, ok)
+		stats, exists := agent.fsStats["nvme0n1"]
+		assert.True(t, exists)
+		assert.True(t, stats.Root)
+		assert.Equal(t, "/", stats.Mountpoint)
+	})
+
+	t.Run("returns false when no io device matches", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{agent: agent, ctx: fsRegistrationContext{diskIoCounters: map[string]disk.IOCountersStat{}}}
+
+		ok := discovery.addPartitionRootFs("/dev/mapper/root", "/")
+
+		assert.False(t, ok)
+		assert.Empty(t, agent.fsStats)
+	})
+}
+
+func TestAddLastResortRootFs(t *testing.T) {
+	t.Run("uses most active io device when available", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{agent: agent, rootMountPoint: "/", ctx: fsRegistrationContext{diskIoCounters: map[string]disk.IOCountersStat{
+			"sda": {Name: "sda", ReadBytes: 5000, WriteBytes: 5000},
+			"sdb": {Name: "sdb", ReadBytes: 1000, WriteBytes: 1000},
+		}}}
+
+		discovery.addLastResortRootFs()
+
+		stats, exists := agent.fsStats["sda"]
+		assert.True(t, exists)
+		assert.True(t, stats.Root)
+	})
+
+	t.Run("falls back to root key when mountpoint basename collides", func(t *testing.T) {
+		agent := &Agent{fsStats: map[string]*system.FsStats{
+			"sysroot": {Mountpoint: "/extra-filesystems/sysroot"},
+		}}
+		discovery := diskDiscovery{agent: agent, rootMountPoint: "/sysroot", ctx: fsRegistrationContext{diskIoCounters: map[string]disk.IOCountersStat{}}}
+
+		discovery.addLastResortRootFs()
+
+		stats, exists := agent.fsStats["root"]
+		assert.True(t, exists)
+		assert.True(t, stats.Root)
+		assert.Equal(t, "/sysroot", stats.Mountpoint)
+	})
+}
+
+func TestAddConfiguredExtraFsEntry(t *testing.T) {
+	t.Run("uses matching partition when present", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent:      agent,
+			partitions: []disk.PartitionStat{{Device: "/dev/sdb1", Mountpoint: "/mnt/backup"}},
+			usageFn: func(string) (*disk.UsageStat, error) {
+				t.Fatal("usage fallback should not be called when partition matches")
+				return nil, nil
+			},
+			ctx: fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"sdb1": {Name: "sdb1"},
+				},
+			},
+		}
+
+		discovery.addConfiguredExtraFsEntry("sdb1", "backup")
+
+		stats, exists := agent.fsStats["sdb1"]
+		assert.True(t, exists)
+		assert.Equal(t, "/mnt/backup", stats.Mountpoint)
+		assert.Equal(t, "backup", stats.Name)
+	})
+
+	t.Run("falls back to usage-validated path", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent: agent,
+			usageFn: func(path string) (*disk.UsageStat, error) {
+				assert.Equal(t, "/srv/archive", path)
+				return &disk.UsageStat{}, nil
+			},
+			ctx: fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"archive": {Name: "archive"},
+				},
+			},
+		}
+
+		discovery.addConfiguredExtraFsEntry("/srv/archive", "archive")
+
+		stats, exists := agent.fsStats["archive"]
+		assert.True(t, exists)
+		assert.Equal(t, "/srv/archive", stats.Mountpoint)
+		assert.Equal(t, "archive", stats.Name)
+	})
+
+	t.Run("ignores invalid filesystem entry", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent: agent,
+			usageFn: func(string) (*disk.UsageStat, error) {
+				return nil, os.ErrNotExist
+			},
+		}
+
+		discovery.addConfiguredExtraFsEntry("/missing/archive", "")
+
+		assert.Empty(t, agent.fsStats)
+	})
+}
+
+func TestAddConfiguredExtraFilesystems(t *testing.T) {
+	t.Run("parses and registers multiple configured filesystems", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		discovery := diskDiscovery{
+			agent:      agent,
+			partitions: []disk.PartitionStat{{Device: "/dev/sda1", Mountpoint: "/mnt/fast"}},
+			usageFn: func(path string) (*disk.UsageStat, error) {
+				if path == "/srv/archive" {
+					return &disk.UsageStat{}, nil
+				}
+				return nil, os.ErrNotExist
+			},
+			ctx: fsRegistrationContext{
+				isWindows: false,
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"sda1":    {Name: "sda1"},
+					"archive": {Name: "archive"},
+				},
+			},
+		}
+
+		discovery.addConfiguredExtraFilesystems("sda1__fast,/srv/archive__cold")
+
+		assert.Contains(t, agent.fsStats, "sda1")
+		assert.Equal(t, "fast", agent.fsStats["sda1"].Name)
+		assert.Contains(t, agent.fsStats, "archive")
+		assert.Equal(t, "cold", agent.fsStats["archive"].Name)
+	})
+}
+
+func TestAddExtraFilesystemFolders(t *testing.T) {
+	t.Run("adds missing folders and skips existing mountpoints", func(t *testing.T) {
+		agent := &Agent{fsStats: map[string]*system.FsStats{
+			"existing": {Mountpoint: "/extra-filesystems/existing"},
+		}}
+		discovery := diskDiscovery{
+			agent: agent,
+			ctx: fsRegistrationContext{
+				isWindows: false,
+				efPath:    "/extra-filesystems",
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"newdisk": {Name: "newdisk"},
+				},
+			},
+		}
+
+		discovery.addExtraFilesystemFolders([]string{"existing", "newdisk__Archive"})
+
+		assert.Len(t, agent.fsStats, 2)
+		stats, exists := agent.fsStats["newdisk"]
+		assert.True(t, exists)
+		assert.Equal(t, "/extra-filesystems/newdisk__Archive", stats.Mountpoint)
+		assert.Equal(t, "Archive", stats.Name)
+	})
+}
+
+func TestAddPartitionExtraFs(t *testing.T) {
+	makeDiscovery := func(agent *Agent) diskDiscovery {
+		return diskDiscovery{
+			agent: agent,
+			ctx: fsRegistrationContext{
+				isWindows: false,
+				efPath:    "/extra-filesystems",
+				diskIoCounters: map[string]disk.IOCountersStat{
+					"nvme0n1p1": {Name: "nvme0n1p1"},
+					"nvme1n1":   {Name: "nvme1n1"},
+				},
+			},
+		}
+	}
+
+	t.Run("registers direct child of extra-filesystems", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		d := makeDiscovery(agent)
+
+		d.addPartitionExtraFs(disk.PartitionStat{
+			Device:     "/dev/nvme0n1p1",
+			Mountpoint: "/extra-filesystems/nvme0n1p1__caddy1-root",
+		})
+
+		stats, exists := agent.fsStats["nvme0n1p1"]
+		assert.True(t, exists)
+		assert.Equal(t, "/extra-filesystems/nvme0n1p1__caddy1-root", stats.Mountpoint)
+		assert.Equal(t, "caddy1-root", stats.Name)
+	})
+
+	t.Run("skips nested mount under extra-filesystem bind mount", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		d := makeDiscovery(agent)
+
+		// These simulate the virtual mounts that appear when host / is bind-mounted
+		// with disk.Partitions(all=true) — e.g. /proc, /sys, /dev visible under the mount.
+		for _, nested := range []string{
+			"/extra-filesystems/nvme0n1p1__caddy1-root/proc",
+			"/extra-filesystems/nvme0n1p1__caddy1-root/sys",
+			"/extra-filesystems/nvme0n1p1__caddy1-root/dev",
+			"/extra-filesystems/nvme0n1p1__caddy1-root/run",
+		} {
+			d.addPartitionExtraFs(disk.PartitionStat{Device: "tmpfs", Mountpoint: nested})
+		}
+
+		assert.Empty(t, agent.fsStats)
+	})
+
+	t.Run("registers both direct children, skips their nested mounts", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		d := makeDiscovery(agent)
+
+		partitions := []disk.PartitionStat{
+			{Device: "/dev/nvme0n1p1", Mountpoint: "/extra-filesystems/nvme0n1p1__caddy1-root"},
+			{Device: "/dev/nvme1n1", Mountpoint: "/extra-filesystems/nvme1n1__caddy1-docker"},
+			{Device: "proc", Mountpoint: "/extra-filesystems/nvme0n1p1__caddy1-root/proc"},
+			{Device: "sysfs", Mountpoint: "/extra-filesystems/nvme0n1p1__caddy1-root/sys"},
+			{Device: "overlay", Mountpoint: "/extra-filesystems/nvme0n1p1__caddy1-root/var/lib/docker"},
+		}
+		for _, p := range partitions {
+			d.addPartitionExtraFs(p)
+		}
+
+		assert.Len(t, agent.fsStats, 2)
+		assert.Equal(t, "caddy1-root", agent.fsStats["nvme0n1p1"].Name)
+		assert.Equal(t, "caddy1-docker", agent.fsStats["nvme1n1"].Name)
+	})
+
+	t.Run("skips partition not under extra-filesystems", func(t *testing.T) {
+		agent := &Agent{fsStats: make(map[string]*system.FsStats)}
+		d := makeDiscovery(agent)
+
+		d.addPartitionExtraFs(disk.PartitionStat{
+			Device:     "/dev/nvme0n1p1",
+			Mountpoint: "/",
+		})
+
+		assert.Empty(t, agent.fsStats)
+	})
+}
+
+func TestFindIoDevice(t *testing.T) {
+	t.Run("matches by device name", func(t *testing.T) {
+		ioCounters := map[string]disk.IOCountersStat{
+			"sda": {Name: "sda"},
+			"sdb": {Name: "sdb"},
+		}
+
+		device, ok := findIoDevice("sdb", ioCounters)
+		assert.True(t, ok)
+		assert.Equal(t, "sdb", device)
+	})
+
+	t.Run("matches by device label", func(t *testing.T) {
+		ioCounters := map[string]disk.IOCountersStat{
+			"sda": {Name: "sda", Label: "rootfs"},
+			"sdb": {Name: "sdb"},
+		}
+
+		device, ok := findIoDevice("rootfs", ioCounters)
+		assert.True(t, ok)
+		assert.Equal(t, "sda", device)
+	})
+
+	t.Run("returns no match when not found", func(t *testing.T) {
+		ioCounters := map[string]disk.IOCountersStat{
+			"sda": {Name: "sda"},
+			"sdb": {Name: "sdb"},
+		}
+
+		device, ok := findIoDevice("nvme0n1p1", ioCounters)
+		assert.False(t, ok)
+		assert.Equal(t, "", device)
+	})
+
+	t.Run("uses uncertain unique prefix fallback", func(t *testing.T) {
+		ioCounters := map[string]disk.IOCountersStat{
+			"nvme0n1": {Name: "nvme0n1"},
+			"sda":     {Name: "sda"},
+		}
+
+		device, ok := findIoDevice("nvme0n1p2", ioCounters)
+		assert.True(t, ok)
+		assert.Equal(t, "nvme0n1", device)
+	})
+
+	t.Run("uses dominant activity when prefix matches are ambiguous", func(t *testing.T) {
+		ioCounters := map[string]disk.IOCountersStat{
+			"sda": {Name: "sda", ReadBytes: 5000, WriteBytes: 5000, ReadCount: 100, WriteCount: 100},
+			"sdb": {Name: "sdb", ReadBytes: 1000, WriteBytes: 1000, ReadCount: 50, WriteCount: 50},
+		}
+
+		device, ok := findIoDevice("sd", ioCounters)
+		assert.True(t, ok)
+		assert.Equal(t, "sda", device)
+	})
+
+	t.Run("uses highest activity when ambiguous without dominance", func(t *testing.T) {
+		ioCounters := map[string]disk.IOCountersStat{
+			"sda": {Name: "sda", ReadBytes: 3000, WriteBytes: 3000, ReadCount: 50, WriteCount: 50},
+			"sdb": {Name: "sdb", ReadBytes: 2500, WriteBytes: 2500, ReadCount: 40, WriteCount: 40},
+		}
+
+		device, ok := findIoDevice("sd", ioCounters)
+		assert.True(t, ok)
+		assert.Equal(t, "sda", device)
+	})
+
+	t.Run("matches /dev/-prefixed partition to parent disk", func(t *testing.T) {
+		ioCounters := map[string]disk.IOCountersStat{
+			"nda0": {Name: "nda0", ReadBytes: 1000, WriteBytes: 1000},
+		}
+
+		device, ok := findIoDevice("/dev/nda0p2", ioCounters)
+		assert.True(t, ok)
+		assert.Equal(t, "nda0", device)
+	})
+
+	t.Run("uses deterministic name tie-breaker", func(t *testing.T) {
+		ioCounters := map[string]disk.IOCountersStat{
+			"sdb": {Name: "sdb", ReadBytes: 2000, WriteBytes: 2000, ReadCount: 10, WriteCount: 10},
+			"sda": {Name: "sda", ReadBytes: 2000, WriteBytes: 2000, ReadCount: 10, WriteCount: 10},
+		}
+
+		device, ok := findIoDevice("sd", ioCounters)
+		assert.True(t, ok)
+		assert.Equal(t, "sda", device)
+	})
+}
+
+func TestFilesystemMatchesPartitionSetting(t *testing.T) {
+	p := disk.PartitionStat{Device: "/dev/ada0p2", Mountpoint: "/"}
+
+	t.Run("matches mountpoint setting", func(t *testing.T) {
+		assert.True(t, filesystemMatchesPartitionSetting("/", p))
+	})
+
+	t.Run("matches exact partition setting", func(t *testing.T) {
+		assert.True(t, filesystemMatchesPartitionSetting("ada0p2", p))
+		assert.True(t, filesystemMatchesPartitionSetting("/dev/ada0p2", p))
+	})
+
+	t.Run("matches prefix-style parent setting", func(t *testing.T) {
+		assert.True(t, filesystemMatchesPartitionSetting("ada0", p))
+		assert.True(t, filesystemMatchesPartitionSetting("/dev/ada0", p))
+	})
+
+	t.Run("does not match unrelated device", func(t *testing.T) {
+		assert.False(t, filesystemMatchesPartitionSetting("sda", p))
+		assert.False(t, filesystemMatchesPartitionSetting("nvme0n1", p))
+		assert.False(t, filesystemMatchesPartitionSetting("", p))
+	})
+}
+
+func TestMostActiveIoDevice(t *testing.T) {
+	t.Run("returns most active device", func(t *testing.T) {
+		ioCounters := map[string]disk.IOCountersStat{
+			"nda0": {Name: "nda0", ReadBytes: 5000, WriteBytes: 5000, ReadCount: 100, WriteCount: 100},
+			"nda1": {Name: "nda1", ReadBytes: 1000, WriteBytes: 1000, ReadCount: 50, WriteCount: 50},
+		}
+		assert.Equal(t, "nda0", mostActiveIoDevice(ioCounters))
+	})
+
+	t.Run("uses deterministic tie-breaker", func(t *testing.T) {
+		ioCounters := map[string]disk.IOCountersStat{
+			"sdb": {Name: "sdb", ReadBytes: 1000, WriteBytes: 1000, ReadCount: 10, WriteCount: 10},
+			"sda": {Name: "sda", ReadBytes: 1000, WriteBytes: 1000, ReadCount: 10, WriteCount: 10},
+		}
+		assert.Equal(t, "sda", mostActiveIoDevice(ioCounters))
+	})
+
+	t.Run("returns empty for empty map", func(t *testing.T) {
+		assert.Equal(t, "", mostActiveIoDevice(map[string]disk.IOCountersStat{}))
+	})
+}
+
+func TestIsDockerSpecialMountpoint(t *testing.T) {
+	testCases := []struct {
+		name       string
+		mountpoint string
+		expected   bool
+	}{
+		{name: "hosts", mountpoint: "/etc/hosts", expected: true},
+		{name: "resolv", mountpoint: "/etc/resolv.conf", expected: true},
+		{name: "hostname", mountpoint: "/etc/hostname", expected: true},
+		{name: "root", mountpoint: "/", expected: false},
+		{name: "passwd", mountpoint: "/etc/passwd", expected: false},
+		{name: "extra-filesystem", mountpoint: "/extra-filesystems/sda1", expected: false},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			assert.Equal(t, tc.expected, isDockerSpecialMountpoint(tc.mountpoint))
+		})
+	}
+}
+
+func TestInitializeDiskInfoWithCustomNames(t *testing.T) {
 	// Test with custom names
-	os.Setenv("EXTRA_FILESYSTEMS", "sda1__my-storage,/dev/sdb1__backup-drive,nvme0n1p2")
+	t.Setenv("EXTRA_FILESYSTEMS", "sda1__my-storage,/dev/sdb1__backup-drive,nvme0n1p2")
 
 	// Mock disk partitions (we'll just test the parsing logic)
 	// Since the actual disk operations are system-dependent, we'll focus on the parsing
@@ -134,7 +797,7 @@ func TestInitializeDiskInfoWithCustomNames(t *testing.T) {
 
 	for _, tc := range testCases {
 		t.Run("env_"+tc.envValue, func(t *testing.T) {
-			os.Setenv("EXTRA_FILESYSTEMS", tc.envValue)
+			t.Setenv("EXTRA_FILESYSTEMS", tc.envValue)
 
 			// Create mock partitions that would match our test cases
 			partitions := []disk.PartitionStat{}
@@ -155,7 +818,7 @@ func TestInitializeDiskInfoWithCustomNames(t *testing.T) {
 			// Test the parsing logic by calling the relevant part
 			// We'll create a simplified version to test just the parsing
 			extraFilesystems := tc.envValue
-			for _, fsEntry := range strings.Split(extraFilesystems, ",") {
+			for fsEntry := range strings.SplitSeq(extraFilesystems, ",") {
 				// Parse the entry
 				fsEntry = strings.TrimSpace(fsEntry)
 				var fs, customName string
@@ -317,3 +980,67 @@ func TestDiskUsageCaching(t *testing.T) {
 			"lastDiskUsageUpdate should be refreshed when cache expires")
 	})
 }
+
+func TestHasSameDiskUsage(t *testing.T) {
+	const toleranceBytes uint64 = 16 * 1024 * 1024
+
+	t.Run("returns true when totals and usage are equal", func(t *testing.T) {
+		a := &disk.UsageStat{Total: 100 * 1024 * 1024 * 1024, Used: 42 * 1024 * 1024 * 1024}
+		b := &disk.UsageStat{Total: 100 * 1024 * 1024 * 1024, Used: 42 * 1024 * 1024 * 1024}
+		assert.True(t, hasSameDiskUsage(a, b))
+	})
+
+	t.Run("returns true within tolerance", func(t *testing.T) {
+		a := &disk.UsageStat{Total: 100 * 1024 * 1024 * 1024, Used: 42 * 1024 * 1024 * 1024}
+		b := &disk.UsageStat{
+			Total: a.Total + toleranceBytes - 1,
+			Used:  a.Used - toleranceBytes + 1,
+		}
+		assert.True(t, hasSameDiskUsage(a, b))
+	})
+
+	t.Run("returns false when total exceeds tolerance", func(t *testing.T) {
+		a := &disk.UsageStat{Total: 100 * 1024 * 1024 * 1024, Used: 42 * 1024 * 1024 * 1024}
+		b := &disk.UsageStat{
+			Total: a.Total + toleranceBytes + 1,
+			Used:  a.Used,
+		}
+		assert.False(t, hasSameDiskUsage(a, b))
+	})
+
+	t.Run("returns false for nil or zero total", func(t *testing.T) {
+		assert.False(t, hasSameDiskUsage(nil, &disk.UsageStat{Total: 1, Used: 1}))
+		assert.False(t, hasSameDiskUsage(&disk.UsageStat{Total: 1, Used: 1}, nil))
+		assert.False(t, hasSameDiskUsage(&disk.UsageStat{Total: 0, Used: 0}, &disk.UsageStat{Total: 1, Used: 1}))
+	})
+}
+
+func TestInitializeDiskIoStatsResetsTrackedDevices(t *testing.T) {
+	agent := &Agent{
+		fsStats: map[string]*system.FsStats{
+			"sda": {},
+			"sdb": {},
+		},
+		fsNames: []string{"stale", "sda"},
+	}
+
+	agent.initializeDiskIoStats(map[string]disk.IOCountersStat{
+		"sda": {Name: "sda", ReadBytes: 10, WriteBytes: 20},
+		"sdb": {Name: "sdb", ReadBytes: 30, WriteBytes: 40},
+	})
+
+	assert.ElementsMatch(t, []string{"sda", "sdb"}, agent.fsNames)
+	assert.Len(t, agent.fsNames, 2)
+	assert.Equal(t, uint64(10), agent.fsStats["sda"].TotalRead)
+	assert.Equal(t, uint64(20), agent.fsStats["sda"].TotalWrite)
+	assert.False(t, agent.fsStats["sda"].Time.IsZero())
+	assert.False(t, agent.fsStats["sdb"].Time.IsZero())
+
+	agent.initializeDiskIoStats(map[string]disk.IOCountersStat{
+		"sdb": {Name: "sdb", ReadBytes: 50, WriteBytes: 60},
+	})
+
+	assert.Equal(t, []string{"sdb"}, agent.fsNames)
+	assert.Equal(t, uint64(50), agent.fsStats["sdb"].TotalRead)
+	assert.Equal(t, uint64(60), agent.fsStats["sdb"].TotalWrite)
+}

agent/docker.go

@@ -1,6 +1,7 @@
 package agent
 
 import (
+	"bufio"
 	"bytes"
 	"context"
 	"encoding/binary"
@@ -15,12 +16,16 @@ import (
 	"os"
 	"path"
 	"regexp"
+	"sort"
+	"strconv"
 	"strings"
 	"sync"
 	"time"
 
 	"github.com/henrygd/beszel/agent/deltatracker"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/container"
+	"github.com/henrygd/beszel/internal/entities/system"
 
 	"github.com/blang/semver"
 )
@@ -28,6 +33,7 @@ import (
 // ansiEscapePattern matches ANSI escape sequences (colors, cursor movement, etc.)
 // This includes CSI sequences like \x1b[...m and simple escapes like \x1b[K
 var ansiEscapePattern = regexp.MustCompile(`\x1b\[[0-9;]*[a-zA-Z]|\x1b\][^\x07]*\x07|\x1b[@-Z\\-_]`)
+var dockerContainerIDPattern = regexp.MustCompile(`^[a-fA-F0-9]{12,64}$`)
 
 const (
 	// Docker API timeout in milliseconds
@@ -47,20 +53,22 @@ const (
 )
 
 type dockerManager struct {
-	client              *http.Client                // Client to query Docker API
-	wg                  sync.WaitGroup              // WaitGroup to wait for all goroutines to finish
-	sem                 chan struct{}               // Semaphore to limit concurrent container requests
-	containerStatsMutex sync.RWMutex                // Mutex to prevent concurrent access to containerStatsMap
-	apiContainerList    []*container.ApiInfo        // List of containers from Docker API
-	containerStatsMap   map[string]*container.Stats // Keeps track of container stats
-	validIds            map[string]struct{}         // Map of valid container ids, used to prune invalid containers from containerStatsMap
-	goodDockerVersion   bool                        // Whether docker version is at least 25.0.0 (one-shot works correctly)
-	isWindows           bool                        // Whether the Docker Engine API is running on Windows
-	buf                 *bytes.Buffer               // Buffer to store and read response bodies
-	decoder             *json.Decoder               // Reusable JSON decoder that reads from buf
-	apiStats            *container.ApiStats         // Reusable API stats object
-	excludeContainers   []string                    // Patterns to exclude containers by name
-	usingPodman         bool                        // Whether the Docker Engine API is running on Podman
+	agent                *Agent                      // Used to propagate system detail changes back to the agent
+	client               *http.Client                // Client to query Docker API
+	wg                   sync.WaitGroup              // WaitGroup to wait for all goroutines to finish
+	sem                  chan struct{}               // Semaphore to limit concurrent container requests
+	containerStatsMutex  sync.RWMutex                // Mutex to prevent concurrent access to containerStatsMap
+	apiContainerList     []*container.ApiInfo        // List of containers from Docker API
+	containerStatsMap    map[string]*container.Stats // Keeps track of container stats
+	validIds             map[string]struct{}         // Map of valid container ids, used to prune invalid containers from containerStatsMap
+	goodDockerVersion    bool                        // Whether docker version is at least 25.0.0 (one-shot works correctly)
+	dockerVersionChecked bool                        // Whether a version probe has completed successfully
+	isWindows            bool                        // Whether the Docker Engine API is running on Windows
+	buf                  *bytes.Buffer               // Buffer to store and read response bodies
+	decoder              *json.Decoder               // Reusable JSON decoder that reads from buf
+	apiStats             *container.ApiStats         // Reusable API stats object
+	excludeContainers    []string                    // Patterns to exclude containers by name
+	usingPodman          bool                        // Whether the Docker Engine API is running on Podman
 
 	// Cache-time-aware tracking for CPU stats (similar to cpu.go)
 	// Maps cache time intervals to container-specific CPU usage tracking
@@ -72,6 +80,7 @@ type dockerManager struct {
 	// cacheTimeMs -> DeltaTracker for network bytes sent/received
 	networkSentTrackers map[uint16]*deltatracker.DeltaTracker[string, uint64]
 	networkRecvTrackers map[uint16]*deltatracker.DeltaTracker[string, uint64]
+	lastNetworkReadTime map[uint16]map[string]time.Time // cacheTimeMs -> containerId -> last network read time
 }
 
 // userAgentRoundTripper is a custom http.RoundTripper that adds a User-Agent header to all requests
@@ -80,6 +89,14 @@ type userAgentRoundTripper struct {
 	userAgent string
 }
 
+// dockerVersionResponse contains the /version fields used for engine checks.
+type dockerVersionResponse struct {
+	Version    string `json:"Version"`
+	Components []struct {
+		Name string `json:"Name"`
+	} `json:"Components"`
+}
+
 // RoundTrip implements the http.RoundTripper interface
 func (u *userAgentRoundTripper) RoundTrip(req *http.Request) (*http.Response, error) {
 	req.Header.Set("User-Agent", u.userAgent)
@@ -127,7 +144,14 @@ func (dm *dockerManager) getDockerStats(cacheTimeMs uint16) ([]*container.Stats,
 		return nil, err
 	}
 
-	dm.isWindows = strings.Contains(resp.Header.Get("Server"), "windows")
+	// Detect Podman and Windows from Server header
+	serverHeader := resp.Header.Get("Server")
+	if !dm.usingPodman && detectPodmanFromHeader(serverHeader) {
+		dm.setIsPodman()
+	}
+	dm.isWindows = strings.Contains(serverHeader, "windows")
+
+	dm.ensureDockerVersionChecked()
 
 	containersLength := len(dm.apiContainerList)
 
@@ -279,7 +303,7 @@ func (dm *dockerManager) cycleNetworkDeltasForCacheTime(cacheTimeMs uint16) {
 }
 
 // calculateNetworkStats calculates network sent/receive deltas using DeltaTracker
-func (dm *dockerManager) calculateNetworkStats(ctr *container.ApiInfo, apiStats *container.ApiStats, stats *container.Stats, initialized bool, name string, cacheTimeMs uint16) (uint64, uint64) {
+func (dm *dockerManager) calculateNetworkStats(ctr *container.ApiInfo, apiStats *container.ApiStats, name string, cacheTimeMs uint16) (uint64, uint64) {
 	var total_sent, total_recv uint64
 	for _, v := range apiStats.Networks {
 		total_sent += v.TxBytes
@@ -298,10 +322,11 @@ func (dm *dockerManager) calculateNetworkStats(ctr *container.ApiInfo, apiStats
 	sent_delta_raw := sentTracker.Delta(ctr.IdShort)
 	recv_delta_raw := recvTracker.Delta(ctr.IdShort)
 
-	// Calculate bytes per second independently for Tx and Rx if we have previous data
+	// Calculate bytes per second using per-cache-time read time to avoid
+	// interference between different cache intervals (e.g. 1000ms vs 60000ms)
 	var sent_delta, recv_delta uint64
-	if initialized {
-		millisecondsElapsed := uint64(time.Since(stats.PrevReadTime).Milliseconds())
+	if prevReadTime, ok := dm.lastNetworkReadTime[cacheTimeMs][ctr.IdShort]; ok {
+		millisecondsElapsed := uint64(time.Since(prevReadTime).Milliseconds())
 		if millisecondsElapsed > 0 {
 			if sent_delta_raw > 0 {
 				sent_delta = sent_delta_raw * 1000 / millisecondsElapsed
@@ -333,15 +358,48 @@ func validateCpuPercentage(cpuPct float64, containerName string) error {
 
 // updateContainerStatsValues updates the final stats values
 func updateContainerStatsValues(stats *container.Stats, cpuPct float64, usedMemory uint64, sent_delta, recv_delta uint64, readTime time.Time) {
-	stats.Cpu = twoDecimals(cpuPct)
-	stats.Mem = bytesToMegabytes(float64(usedMemory))
+	stats.Cpu = utils.TwoDecimals(cpuPct)
+	stats.Mem = utils.BytesToMegabytes(float64(usedMemory))
 	stats.Bandwidth = [2]uint64{sent_delta, recv_delta}
 	// TODO(0.19+): stop populating NetworkSent/NetworkRecv (deprecated in 0.18.3)
-	stats.NetworkSent = bytesToMegabytes(float64(sent_delta))
-	stats.NetworkRecv = bytesToMegabytes(float64(recv_delta))
+	stats.NetworkSent = utils.BytesToMegabytes(float64(sent_delta))
+	stats.NetworkRecv = utils.BytesToMegabytes(float64(recv_delta))
 	stats.PrevReadTime = readTime
 }
 
+// convertContainerPortsToString formats the ports of a container into a sorted, deduplicated string.
+// ctr.Ports is nilled out after processing so the slice is not accidentally reused.
+func convertContainerPortsToString(ctr *container.ApiInfo) string {
+	if len(ctr.Ports) == 0 {
+		return ""
+	}
+	sort.Slice(ctr.Ports, func(i, j int) bool {
+		return ctr.Ports[i].PublicPort < ctr.Ports[j].PublicPort
+	})
+	var builder strings.Builder
+	seenPorts := make(map[uint16]struct{})
+	for _, p := range ctr.Ports {
+		_, ok := seenPorts[p.PublicPort]
+		if p.PublicPort == 0 || ok {
+			continue
+		}
+		seenPorts[p.PublicPort] = struct{}{}
+		if builder.Len() > 0 {
+			builder.WriteString(", ")
+		}
+		switch p.IP {
+		case "0.0.0.0", "::":
+		default:
+			builder.WriteString(p.IP)
+			builder.WriteByte(':')
+		}
+		builder.WriteString(strconv.Itoa(int(p.PublicPort)))
+	}
+	// clear ports slice so it doesn't get reused and blend into next response
+	ctr.Ports = nil
+	return builder.String()
+}
+
 func parseDockerStatus(status string) (string, container.DockerHealth) {
 	trimmed := strings.TrimSpace(status)
 	if trimmed == "" {
@@ -361,22 +419,60 @@ func parseDockerStatus(status string) (string, container.DockerHealth) {
 		statusText = trimmed
 	}
 
-	healthText := strings.ToLower(strings.TrimSpace(strings.TrimSuffix(trimmed[openIdx+1:], ")")))
+	healthText := strings.TrimSpace(strings.TrimSuffix(trimmed[openIdx+1:], ")"))
 	// Some Docker statuses include a "health:" prefix inside the parentheses.
 	// Strip it so it maps correctly to the known health states.
 	if colonIdx := strings.IndexRune(healthText, ':'); colonIdx != -1 {
-		prefix := strings.TrimSpace(healthText[:colonIdx])
+		prefix := strings.ToLower(strings.TrimSpace(healthText[:colonIdx]))
 		if prefix == "health" || prefix == "health status" {
 			healthText = strings.TrimSpace(healthText[colonIdx+1:])
 		}
 	}
-	if health, ok := container.DockerHealthStrings[healthText]; ok {
+	if health, ok := parseDockerHealthStatus(healthText); ok {
 		return statusText, health
 	}
 
 	return trimmed, container.DockerHealthNone
 }
 
+// parseDockerHealthStatus maps Docker health status strings to container.DockerHealth values
+func parseDockerHealthStatus(status string) (container.DockerHealth, bool) {
+	health, ok := container.DockerHealthStrings[strings.ToLower(strings.TrimSpace(status))]
+	return health, ok
+}
+
+// getPodmanContainerHealth fetches container health status from the container inspect endpoint.
+// Used for Podman which doesn't provide health status in the /containers/json endpoint as of March 2026.
+// https://github.com/containers/podman/issues/27786
+func (dm *dockerManager) getPodmanContainerHealth(containerID string) (container.DockerHealth, error) {
+	resp, err := dm.client.Get(fmt.Sprintf("http://localhost/containers/%s/json", url.PathEscape(containerID)))
+	if err != nil {
+		return container.DockerHealthNone, err
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		return container.DockerHealthNone, fmt.Errorf("container inspect request failed: %s", resp.Status)
+	}
+
+	var inspectInfo struct {
+		State struct {
+			Health struct {
+				Status string
+			}
+		}
+	}
+	if err := json.NewDecoder(resp.Body).Decode(&inspectInfo); err != nil {
+		return container.DockerHealthNone, err
+	}
+
+	if health, ok := parseDockerHealthStatus(inspectInfo.State.Health.Status); ok {
+		return health, nil
+	}
+
+	return container.DockerHealthNone, nil
+}
+
 // Updates stats for individual container with cache-time-aware delta tracking
 func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo, cacheTimeMs uint16) error {
 	name := ctr.Names[0][1:]
@@ -386,6 +482,21 @@ func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo, cacheTimeM
 		return err
 	}
 
+	statusText, health := parseDockerStatus(ctr.Status)
+
+	// Docker exposes Health.Status on /containers/json in API 1.52+.
+	// Podman currently requires falling back to the inspect endpoint as of March 2026.
+	// https://github.com/containers/podman/issues/27786
+	if ctr.Health.Status != "" {
+		if h, ok := parseDockerHealthStatus(ctr.Health.Status); ok {
+			health = h
+		}
+	} else if dm.usingPodman {
+		if podmanHealth, err := dm.getPodmanContainerHealth(ctr.IdShort); err == nil {
+			health = podmanHealth
+		}
+	}
+
 	dm.containerStatsMutex.Lock()
 	defer dm.containerStatsMutex.Unlock()
 
@@ -397,11 +508,13 @@ func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo, cacheTimeM
 	}
 
 	stats.Id = ctr.IdShort
-
-	statusText, health := parseDockerStatus(ctr.Status)
 	stats.Status = statusText
 	stats.Health = health
 
+	if len(ctr.Ports) > 0 {
+		stats.Ports = convertContainerPortsToString(ctr)
+	}
+
 	// reset current stats
 	stats.Cpu = 0
 	stats.Mem = 0
@@ -448,7 +561,13 @@ func (dm *dockerManager) updateContainerStats(ctr *container.ApiInfo, cacheTimeM
 	}
 
 	// Calculate network stats using DeltaTracker
-	sent_delta, recv_delta := dm.calculateNetworkStats(ctr, res, stats, initialized, name, cacheTimeMs)
+	sent_delta, recv_delta := dm.calculateNetworkStats(ctr, res, name, cacheTimeMs)
+
+	// Store per-cache-time network read time for next rate calculation
+	if dm.lastNetworkReadTime[cacheTimeMs] == nil {
+		dm.lastNetworkReadTime[cacheTimeMs] = make(map[string]time.Time)
+	}
+	dm.lastNetworkReadTime[cacheTimeMs][ctr.IdShort] = time.Now()
 
 	// Store current network values for legacy compatibility
 	var total_sent, total_recv uint64
@@ -480,11 +599,14 @@ func (dm *dockerManager) deleteContainerStatsSync(id string) {
 	for ct := range dm.lastCpuReadTime {
 		delete(dm.lastCpuReadTime[ct], id)
 	}
+	for ct := range dm.lastNetworkReadTime {
+		delete(dm.lastNetworkReadTime[ct], id)
+	}
 }
 
 // Creates a new http client for Docker or Podman API
-func newDockerManager() *dockerManager {
-	dockerHost, exists := GetEnv("DOCKER_HOST")
+func newDockerManager(agent *Agent) *dockerManager {
+	dockerHost, exists := utils.GetEnv("DOCKER_HOST")
 	if exists {
 		// return nil if set to empty string
 		if dockerHost == "" {
@@ -520,7 +642,7 @@ func newDockerManager() *dockerManager {
 
 	// configurable timeout
 	timeout := time.Millisecond * time.Duration(dockerTimeoutMs)
-	if t, set := GetEnv("DOCKER_TIMEOUT"); set {
+	if t, set := utils.GetEnv("DOCKER_TIMEOUT"); set {
 		timeout, err = time.ParseDuration(t)
 		if err != nil {
 			slog.Error(err.Error())
@@ -537,7 +659,7 @@ func newDockerManager() *dockerManager {
 
 	// Read container exclusion patterns from environment variable
 	var excludeContainers []string
-	if excludeStr, set := GetEnv("EXCLUDE_CONTAINERS"); set && excludeStr != "" {
+	if excludeStr, set := utils.GetEnv("EXCLUDE_CONTAINERS"); set && excludeStr != "" {
 		parts := strings.SplitSeq(excludeStr, ",")
 		for part := range parts {
 			trimmed := strings.TrimSpace(part)
@@ -549,6 +671,7 @@ func newDockerManager() *dockerManager {
 	}
 
 	manager := &dockerManager{
+		agent: agent,
 		client: &http.Client{
 			Timeout:   timeout,
 			Transport: userAgentTransport,
@@ -565,50 +688,55 @@ func newDockerManager() *dockerManager {
 		lastCpuReadTime:     make(map[uint16]map[string]time.Time),
 		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
 		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		lastNetworkReadTime: make(map[uint16]map[string]time.Time),
 	}
 
-	// If using podman, return client
-	if strings.Contains(dockerHost, "podman") {
-		manager.usingPodman = true
-		manager.goodDockerVersion = true
-		return manager
-	}
-
-	// this can take up to 5 seconds with retry, so run in goroutine
-	go manager.checkDockerVersion()
-
-	// give version check a chance to complete before returning
-	time.Sleep(50 * time.Millisecond)
+	// Best-effort startup probe. If the engine is not ready yet, getDockerStats will
+	// retry after the first successful /containers/json request.
+	_, _ = manager.checkDockerVersion()
 
 	return manager
 }
 
 // checkDockerVersion checks Docker version and sets goodDockerVersion if at least 25.0.0.
 // Versions before 25.0.0 have a bug with one-shot which requires all requests to be made in one batch.
-func (dm *dockerManager) checkDockerVersion() {
-	var err error
-	var resp *http.Response
-	var versionInfo struct {
-		Version string `json:"Version"`
-	}
-	const versionMaxTries = 2
-	for i := 1; i <= versionMaxTries; i++ {
-		resp, err = dm.client.Get("http://localhost/version")
-		if err == nil {
-			break
-		}
-		if resp != nil {
-			resp.Body.Close()
-		}
-		if i < versionMaxTries {
-			slog.Debug("Failed to get Docker version; retrying", "attempt", i, "error", err)
-			time.Sleep(5 * time.Second)
-		}
-	}
+func (dm *dockerManager) checkDockerVersion() (bool, error) {
+	resp, err := dm.client.Get("http://localhost/version")
 	if err != nil {
+		return false, err
+	}
+	if resp.StatusCode != http.StatusOK {
+		status := resp.Status
+		resp.Body.Close()
+		return false, fmt.Errorf("docker version request failed: %s", status)
+	}
+
+	var versionInfo dockerVersionResponse
+	serverHeader := resp.Header.Get("Server")
+	if err := dm.decode(resp, &versionInfo); err != nil {
+		return false, err
+	}
+
+	dm.applyDockerVersionInfo(serverHeader, &versionInfo)
+	dm.dockerVersionChecked = true
+	return true, nil
+}
+
+// ensureDockerVersionChecked retries the version probe after a successful
+// container list request.
+func (dm *dockerManager) ensureDockerVersionChecked() {
+	if dm.dockerVersionChecked {
 		return
 	}
-	if err := dm.decode(resp, &versionInfo); err != nil {
+	if _, err := dm.checkDockerVersion(); err != nil {
+		slog.Debug("Failed to get Docker version", "err", err)
+	}
+}
+
+// applyDockerVersionInfo updates version-dependent behavior from engine metadata.
+func (dm *dockerManager) applyDockerVersionInfo(serverHeader string, versionInfo *dockerVersionResponse) {
+	if detectPodmanEngine(serverHeader, versionInfo) {
+		dm.setIsPodman()
 		return
 	}
 	// if version > 24, one-shot works correctly and we can limit concurrent operations
@@ -647,9 +775,34 @@ func getDockerHost() string {
 	return scheme + socks[0]
 }
 
+func validateContainerID(containerID string) error {
+	if !dockerContainerIDPattern.MatchString(containerID) {
+		return fmt.Errorf("invalid container id")
+	}
+	return nil
+}
+
+func buildDockerContainerEndpoint(containerID, action string, query url.Values) (string, error) {
+	if err := validateContainerID(containerID); err != nil {
+		return "", err
+	}
+	u := &url.URL{
+		Scheme: "http",
+		Host:   "localhost",
+		Path:   fmt.Sprintf("/containers/%s/%s", url.PathEscape(containerID), action),
+	}
+	if len(query) > 0 {
+		u.RawQuery = query.Encode()
+	}
+	return u.String(), nil
+}
+
 // getContainerInfo fetches the inspection data for a container
 func (dm *dockerManager) getContainerInfo(ctx context.Context, containerID string) ([]byte, error) {
-	endpoint := fmt.Sprintf("http://localhost/containers/%s/json", containerID)
+	endpoint, err := buildDockerContainerEndpoint(containerID, "json", nil)
+	if err != nil {
+		return nil, err
+	}
 	req, err := http.NewRequestWithContext(ctx, http.MethodGet, endpoint, nil)
 	if err != nil {
 		return nil, err
@@ -680,7 +833,15 @@ func (dm *dockerManager) getContainerInfo(ctx context.Context, containerID strin
 
 // getLogs fetches the logs for a container
 func (dm *dockerManager) getLogs(ctx context.Context, containerID string) (string, error) {
-	endpoint := fmt.Sprintf("http://localhost/containers/%s/logs?stdout=1&stderr=1&tail=%d", containerID, dockerLogsTail)
+	query := url.Values{
+		"stdout": []string{"1"},
+		"stderr": []string{"1"},
+		"tail":   []string{fmt.Sprintf("%d", dockerLogsTail)},
+	}
+	endpoint, err := buildDockerContainerEndpoint(containerID, "logs", query)
+	if err != nil {
+		return "", err
+	}
 	req, err := http.NewRequestWithContext(ctx, http.MethodGet, endpoint, nil)
 	if err != nil {
 		return "", err
@@ -698,8 +859,17 @@ func (dm *dockerManager) getLogs(ctx context.Context, containerID string) (strin
 	}
 
 	var builder strings.Builder
-	multiplexed := resp.Header.Get("Content-Type") == "application/vnd.docker.multiplexed-stream"
-	if err := decodeDockerLogStream(resp.Body, &builder, multiplexed); err != nil {
+	contentType := resp.Header.Get("Content-Type")
+	multiplexed := strings.HasSuffix(contentType, "multiplexed-stream")
+	logReader := io.Reader(resp.Body)
+	if !multiplexed {
+		// Podman may return multiplexed logs without Content-Type. Sniff the first frame header
+		// with a small buffered reader only when the header check fails.
+		bufferedReader := bufio.NewReaderSize(resp.Body, 8)
+		multiplexed = detectDockerMultiplexedStream(bufferedReader)
+		logReader = bufferedReader
+	}
+	if err := decodeDockerLogStream(logReader, &builder, multiplexed); err != nil {
 		return "", err
 	}
 
@@ -711,6 +881,23 @@ func (dm *dockerManager) getLogs(ctx context.Context, containerID string) (strin
 	return logs, nil
 }
 
+func detectDockerMultiplexedStream(reader *bufio.Reader) bool {
+	const headerSize = 8
+	header, err := reader.Peek(headerSize)
+	if err != nil {
+		return false
+	}
+	if header[0] != 0x01 && header[0] != 0x02 {
+		return false
+	}
+	// Docker's stream framing header reserves bytes 1-3 as zero.
+	if header[1] != 0 || header[2] != 0 || header[3] != 0 {
+		return false
+	}
+	frameLen := binary.BigEndian.Uint32(header[4:])
+	return frameLen <= maxLogFrameSize
+}
+
 func decodeDockerLogStream(reader io.Reader, builder *strings.Builder, multiplexed bool) error {
 	if !multiplexed {
 		_, err := io.Copy(builder, io.LimitReader(reader, maxTotalLogSize))
@@ -775,3 +962,46 @@ func (dm *dockerManager) GetHostInfo() (info container.HostInfo, err error) {
 func (dm *dockerManager) IsPodman() bool {
 	return dm.usingPodman
 }
+
+// setIsPodman sets the manager to Podman mode and updates system details accordingly.
+func (dm *dockerManager) setIsPodman() {
+	if dm.usingPodman {
+		return
+	}
+	dm.usingPodman = true
+	dm.goodDockerVersion = true
+	dm.dockerVersionChecked = true
+	// keep system details updated - this may be detected late if server isn't ready when
+	// agent starts, so make sure we notify the hub if this happens later.
+	if dm.agent != nil {
+		dm.agent.updateSystemDetails(func(details *system.Details) {
+			details.Podman = true
+		})
+	}
+}
+
+// detectPodmanFromHeader identifies Podman from the Docker API server header.
+func detectPodmanFromHeader(server string) bool {
+	return strings.HasPrefix(server, "Libpod")
+}
+
+// detectPodmanFromVersion identifies Podman from the version payload.
+func detectPodmanFromVersion(versionInfo *dockerVersionResponse) bool {
+	if versionInfo == nil {
+		return false
+	}
+	for _, component := range versionInfo.Components {
+		if strings.HasPrefix(component.Name, "Podman") {
+			return true
+		}
+	}
+	return false
+}
+
+// detectPodmanEngine checks both header and version metadata for Podman.
+func detectPodmanEngine(serverHeader string, versionInfo *dockerVersionResponse) bool {
+	if detectPodmanFromHeader(serverHeader) {
+		return true
+	}
+	return detectPodmanFromVersion(versionInfo)
+}

agent/docker_test.go

@@ -1,17 +1,24 @@
 //go:build testing
-// +build testing
 
 package agent
 
 import (
 	"bytes"
+	"context"
 	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"net"
+	"net/http"
+	"net/http/httptest"
 	"os"
 	"strings"
 	"testing"
 	"time"
 
 	"github.com/henrygd/beszel/agent/deltatracker"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/container"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
@@ -19,6 +26,43 @@ import (
 
 var defaultCacheTimeMs = uint16(60_000)
 
+type recordingRoundTripper struct {
+	statusCode  int
+	body        string
+	contentType string
+	called      bool
+	lastPath    string
+	lastQuery   map[string]string
+}
+
+type roundTripFunc func(*http.Request) (*http.Response, error)
+
+func (fn roundTripFunc) RoundTrip(req *http.Request) (*http.Response, error) {
+	return fn(req)
+}
+
+func (rt *recordingRoundTripper) RoundTrip(req *http.Request) (*http.Response, error) {
+	rt.called = true
+	rt.lastPath = req.URL.EscapedPath()
+	rt.lastQuery = map[string]string{}
+	for key, values := range req.URL.Query() {
+		if len(values) > 0 {
+			rt.lastQuery[key] = values[0]
+		}
+	}
+	resp := &http.Response{
+		StatusCode: rt.statusCode,
+		Status:     "200 OK",
+		Header:     make(http.Header),
+		Body:       io.NopCloser(strings.NewReader(rt.body)),
+		Request:    req,
+	}
+	if rt.contentType != "" {
+		resp.Header.Set("Content-Type", rt.contentType)
+	}
+	return resp, nil
+}
+
 // cycleCpuDeltas cycles the CPU tracking data for a specific cache time interval
 func (dm *dockerManager) cycleCpuDeltas(cacheTimeMs uint16) {
 	// Clear the CPU tracking maps for this cache time interval
@@ -110,6 +154,94 @@ func TestCalculateMemoryUsage(t *testing.T) {
 	}
 }
 
+func TestBuildDockerContainerEndpoint(t *testing.T) {
+	t.Run("valid container ID builds escaped endpoint", func(t *testing.T) {
+		endpoint, err := buildDockerContainerEndpoint("0123456789ab", "json", nil)
+		require.NoError(t, err)
+		assert.Equal(t, "http://localhost/containers/0123456789ab/json", endpoint)
+	})
+
+	t.Run("invalid container ID is rejected", func(t *testing.T) {
+		_, err := buildDockerContainerEndpoint("../../version", "json", nil)
+		require.Error(t, err)
+		assert.Contains(t, err.Error(), "invalid container id")
+	})
+}
+
+func TestContainerDetailsRequestsValidateContainerID(t *testing.T) {
+	rt := &recordingRoundTripper{
+		statusCode: 200,
+		body:       `{"Config":{"Env":["SECRET=1"]}}`,
+	}
+	dm := &dockerManager{
+		client: &http.Client{Transport: rt},
+	}
+
+	_, err := dm.getContainerInfo(context.Background(), "../version")
+	require.Error(t, err)
+	assert.Contains(t, err.Error(), "invalid container id")
+	assert.False(t, rt.called, "request should be rejected before dispatching to Docker API")
+}
+
+func TestContainerDetailsRequestsUseExpectedDockerPaths(t *testing.T) {
+	t.Run("container info uses container json endpoint", func(t *testing.T) {
+		rt := &recordingRoundTripper{
+			statusCode: 200,
+			body:       `{"Config":{"Env":["SECRET=1"]},"Name":"demo"}`,
+		}
+		dm := &dockerManager{
+			client: &http.Client{Transport: rt},
+		}
+
+		body, err := dm.getContainerInfo(context.Background(), "0123456789ab")
+		require.NoError(t, err)
+		assert.True(t, rt.called)
+		assert.Equal(t, "/containers/0123456789ab/json", rt.lastPath)
+		assert.NotContains(t, string(body), "SECRET=1", "sensitive env vars should be removed")
+	})
+
+	t.Run("container logs uses expected endpoint and query params", func(t *testing.T) {
+		rt := &recordingRoundTripper{
+			statusCode: 200,
+			body:       "line1\nline2\n",
+		}
+		dm := &dockerManager{
+			client: &http.Client{Transport: rt},
+		}
+
+		logs, err := dm.getLogs(context.Background(), "abcdef123456")
+		require.NoError(t, err)
+		assert.True(t, rt.called)
+		assert.Equal(t, "/containers/abcdef123456/logs", rt.lastPath)
+		assert.Equal(t, "1", rt.lastQuery["stdout"])
+		assert.Equal(t, "1", rt.lastQuery["stderr"])
+		assert.Equal(t, "200", rt.lastQuery["tail"])
+		assert.Equal(t, "line1\nline2\n", logs)
+	})
+}
+
+func TestGetPodmanContainerHealth(t *testing.T) {
+	called := false
+	dm := &dockerManager{
+		client: &http.Client{Transport: roundTripFunc(func(req *http.Request) (*http.Response, error) {
+			called = true
+			assert.Equal(t, "/containers/0123456789ab/json", req.URL.EscapedPath())
+			return &http.Response{
+				StatusCode: http.StatusOK,
+				Status:     "200 OK",
+				Header:     make(http.Header),
+				Body:       io.NopCloser(strings.NewReader(`{"State":{"Health":{"Status":"healthy"}}}`)),
+				Request:    req,
+			}, nil
+		})},
+	}
+
+	health, err := dm.getPodmanContainerHealth("0123456789ab")
+	require.NoError(t, err)
+	assert.True(t, called)
+	assert.Equal(t, container.DockerHealthHealthy, health)
+}
+
 func TestValidateCpuPercentage(t *testing.T) {
 	tests := []struct {
 		name          string
@@ -195,48 +327,6 @@ func TestUpdateContainerStatsValues(t *testing.T) {
 	assert.Equal(t, testTime, stats.PrevReadTime)
 }
 
-func TestTwoDecimals(t *testing.T) {
-	tests := []struct {
-		name     string
-		input    float64
-		expected float64
-	}{
-		{"round down", 1.234, 1.23},
-		{"round half up", 1.235, 1.24}, // math.Round rounds half up
-		{"no rounding needed", 1.23, 1.23},
-		{"negative number", -1.235, -1.24}, // math.Round rounds half up (more negative)
-		{"zero", 0.0, 0.0},
-		{"large number", 123.456, 123.46}, // rounds 5 up
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			result := twoDecimals(tt.input)
-			assert.Equal(t, tt.expected, result)
-		})
-	}
-}
-
-func TestBytesToMegabytes(t *testing.T) {
-	tests := []struct {
-		name     string
-		input    float64
-		expected float64
-	}{
-		{"1 MB", 1048576, 1.0},
-		{"512 KB", 524288, 0.5},
-		{"zero", 0, 0},
-		{"large value", 1073741824, 1024}, // 1 GB = 1024 MB
-	}
-
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			result := bytesToMegabytes(tt.input)
-			assert.Equal(t, tt.expected, result)
-		})
-	}
-}
-
 func TestInitializeCpuTracking(t *testing.T) {
 	dm := &dockerManager{
 		lastCpuContainer: make(map[uint16]map[string]uint64),
@@ -318,6 +408,7 @@ func TestCalculateNetworkStats(t *testing.T) {
 	dm := &dockerManager{
 		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
 		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		lastNetworkReadTime: make(map[uint16]map[string]time.Time),
 	}
 
 	cacheTimeMs := uint16(30000)
@@ -333,6 +424,11 @@ func TestCalculateNetworkStats(t *testing.T) {
 	dm.networkSentTrackers[cacheTimeMs] = sentTracker
 	dm.networkRecvTrackers[cacheTimeMs] = recvTracker
 
+	// Set per-cache-time network read time (1 second ago)
+	dm.lastNetworkReadTime[cacheTimeMs] = map[string]time.Time{
+		"container1": time.Now().Add(-time.Second),
+	}
+
 	ctr := &container.ApiInfo{
 		IdShort: "container1",
 	}
@@ -343,12 +439,8 @@ func TestCalculateNetworkStats(t *testing.T) {
 		},
 	}
 
-	stats := &container.Stats{
-		PrevReadTime: time.Now().Add(-time.Second), // 1 second ago
-	}
-
 	// Test with initialized container
-	sent, recv := dm.calculateNetworkStats(ctr, apiStats, stats, true, "test-container", cacheTimeMs)
+	sent, recv := dm.calculateNetworkStats(ctr, apiStats, "test-container", cacheTimeMs)
 
 	// Should return calculated byte rates per second
 	assert.GreaterOrEqual(t, sent, uint64(0))
@@ -356,12 +448,76 @@ func TestCalculateNetworkStats(t *testing.T) {
 
 	// Cycle and test one-direction change (Tx only) is reflected independently
 	dm.cycleNetworkDeltasForCacheTime(cacheTimeMs)
+	dm.lastNetworkReadTime[cacheTimeMs]["container1"] = time.Now().Add(-time.Second)
 	apiStats.Networks["eth0"] = container.NetworkStats{TxBytes: 2500, RxBytes: 1800} // +500 Tx only
-	sent, recv = dm.calculateNetworkStats(ctr, apiStats, stats, true, "test-container", cacheTimeMs)
+	sent, recv = dm.calculateNetworkStats(ctr, apiStats, "test-container", cacheTimeMs)
 	assert.Greater(t, sent, uint64(0))
 	assert.Equal(t, uint64(0), recv)
 }
 
+// TestNetworkStatsCacheTimeIsolation verifies that frequent collections at one cache time
+// (e.g. 1000ms) don't cause inflated rates at another cache time (e.g. 60000ms).
+// This was a bug where PrevReadTime was shared, so the 60000ms tracker would see a
+// large byte delta divided by a tiny elapsed time (set by the 1000ms path).
+func TestNetworkStatsCacheTimeIsolation(t *testing.T) {
+	dm := &dockerManager{
+		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		lastNetworkReadTime: make(map[uint16]map[string]time.Time),
+	}
+
+	ctr := &container.ApiInfo{IdShort: "container1"}
+	fastCache := uint16(1000)
+	slowCache := uint16(60000)
+
+	// Baseline for both cache times at T=0 with 100 bytes total
+	baseline := &container.ApiStats{
+		Networks: map[string]container.NetworkStats{
+			"eth0": {TxBytes: 100, RxBytes: 100},
+		},
+	}
+	dm.calculateNetworkStats(ctr, baseline, "test", fastCache)
+	dm.calculateNetworkStats(ctr, baseline, "test", slowCache)
+
+	// Record read times and cycle both
+	now := time.Now()
+	dm.lastNetworkReadTime[fastCache] = map[string]time.Time{"container1": now}
+	dm.lastNetworkReadTime[slowCache] = map[string]time.Time{"container1": now}
+	dm.cycleNetworkDeltasForCacheTime(fastCache)
+	dm.cycleNetworkDeltasForCacheTime(slowCache)
+
+	// Simulate many fast (1000ms) collections over ~5 seconds, each adding 10 bytes
+	totalBytes := uint64(100)
+	for i := 0; i < 5; i++ {
+		totalBytes += 10
+		stats := &container.ApiStats{
+			Networks: map[string]container.NetworkStats{
+				"eth0": {TxBytes: totalBytes, RxBytes: totalBytes},
+			},
+		}
+		// Set fast cache read time to 1 second ago
+		dm.lastNetworkReadTime[fastCache]["container1"] = time.Now().Add(-time.Second)
+		sent, _ := dm.calculateNetworkStats(ctr, stats, "test", fastCache)
+		// Fast cache should see ~10 bytes/sec per interval
+		assert.LessOrEqual(t, sent, uint64(100), "fast cache rate should be reasonable")
+		dm.cycleNetworkDeltasForCacheTime(fastCache)
+	}
+
+	// Now do slow cache collection — total delta is 50 bytes over ~5 seconds
+	// Set slow cache read time to 5 seconds ago (the actual elapsed time)
+	dm.lastNetworkReadTime[slowCache]["container1"] = time.Now().Add(-5 * time.Second)
+	finalStats := &container.ApiStats{
+		Networks: map[string]container.NetworkStats{
+			"eth0": {TxBytes: totalBytes, RxBytes: totalBytes},
+		},
+	}
+	sent, _ := dm.calculateNetworkStats(ctr, finalStats, "test", slowCache)
+
+	// Slow cache rate should be ~10 bytes/sec (50 bytes / 5 seconds), NOT 100x inflated
+	assert.LessOrEqual(t, sent, uint64(100), "slow cache rate should NOT be inflated by fast cache collections")
+	assert.GreaterOrEqual(t, sent, uint64(1), "slow cache should still report some traffic")
+}
+
 func TestDockerManagerCreation(t *testing.T) {
 	// Test that dockerManager can be created without panicking
 	dm := &dockerManager{
@@ -370,6 +526,7 @@ func TestDockerManagerCreation(t *testing.T) {
 		lastCpuReadTime:     make(map[uint16]map[string]time.Time),
 		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
 		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		lastNetworkReadTime: make(map[uint16]map[string]time.Time),
 	}
 
 	assert.NotNil(t, dm)
@@ -377,6 +534,274 @@ func TestDockerManagerCreation(t *testing.T) {
 	assert.NotNil(t, dm.lastCpuSystem)
 	assert.NotNil(t, dm.networkSentTrackers)
 	assert.NotNil(t, dm.networkRecvTrackers)
+	assert.NotNil(t, dm.lastNetworkReadTime)
+}
+
+func TestCheckDockerVersion(t *testing.T) {
+	tests := []struct {
+		name            string
+		statusCode      int
+		body            string
+		server          string
+		expectSuccess   bool
+		expectedGood    bool
+		expectedPodman  bool
+		expectError     bool
+		expectedRequest string
+	}{
+		{
+			name:            "good docker version",
+			statusCode:      http.StatusOK,
+			body:            `{"Version":"25.0.1"}`,
+			expectSuccess:   true,
+			expectedGood:    true,
+			expectedPodman:  false,
+			expectedRequest: "/version",
+		},
+		{
+			name:            "old docker version",
+			statusCode:      http.StatusOK,
+			body:            `{"Version":"24.0.7"}`,
+			expectSuccess:   true,
+			expectedGood:    false,
+			expectedPodman:  false,
+			expectedRequest: "/version",
+		},
+		{
+			name:            "podman from server header",
+			statusCode:      http.StatusOK,
+			body:            `{"Version":"5.5.0"}`,
+			server:          "Libpod/5.5.0",
+			expectSuccess:   true,
+			expectedGood:    true,
+			expectedPodman:  true,
+			expectedRequest: "/version",
+		},
+		{
+			name:            "non-200 response",
+			statusCode:      http.StatusServiceUnavailable,
+			body:            `"not ready"`,
+			expectSuccess:   false,
+			expectedGood:    false,
+			expectedPodman:  false,
+			expectError:     true,
+			expectedRequest: "/version",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			requestCount := 0
+			server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				requestCount++
+				assert.Equal(t, tt.expectedRequest, r.URL.EscapedPath())
+				if tt.server != "" {
+					w.Header().Set("Server", tt.server)
+				}
+				w.WriteHeader(tt.statusCode)
+				fmt.Fprint(w, tt.body)
+			}))
+			defer server.Close()
+
+			dm := &dockerManager{
+				client: &http.Client{
+					Transport: &http.Transport{
+						DialContext: func(_ context.Context, network, _ string) (net.Conn, error) {
+							return net.Dial(network, server.Listener.Addr().String())
+						},
+					},
+				},
+			}
+
+			success, err := dm.checkDockerVersion()
+
+			assert.Equal(t, tt.expectSuccess, success)
+			assert.Equal(t, tt.expectSuccess, dm.dockerVersionChecked)
+			assert.Equal(t, tt.expectedGood, dm.goodDockerVersion)
+			assert.Equal(t, tt.expectedPodman, dm.usingPodman)
+			assert.Equal(t, 1, requestCount)
+			if tt.expectError {
+				require.Error(t, err)
+			} else {
+				require.NoError(t, err)
+			}
+		})
+	}
+
+	t.Run("request error", func(t *testing.T) {
+		requestCount := 0
+		dm := &dockerManager{
+			client: &http.Client{
+				Transport: &http.Transport{
+					DialContext: func(_ context.Context, _, _ string) (net.Conn, error) {
+						requestCount++
+						return nil, errors.New("connection refused")
+					},
+				},
+			},
+		}
+
+		success, err := dm.checkDockerVersion()
+
+		assert.False(t, success)
+		require.Error(t, err)
+		assert.False(t, dm.dockerVersionChecked)
+		assert.False(t, dm.goodDockerVersion)
+		assert.False(t, dm.usingPodman)
+		assert.Equal(t, 1, requestCount)
+	})
+}
+
+// newDockerManagerForVersionTest creates a dockerManager wired to a test server.
+func newDockerManagerForVersionTest(server *httptest.Server) *dockerManager {
+	return &dockerManager{
+		client: &http.Client{
+			Transport: &http.Transport{
+				DialContext: func(_ context.Context, network, _ string) (net.Conn, error) {
+					return net.Dial(network, server.Listener.Addr().String())
+				},
+			},
+		},
+		containerStatsMap:   make(map[string]*container.Stats),
+		lastCpuContainer:    make(map[uint16]map[string]uint64),
+		lastCpuSystem:       make(map[uint16]map[string]uint64),
+		lastCpuReadTime:     make(map[uint16]map[string]time.Time),
+		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		lastNetworkReadTime: make(map[uint16]map[string]time.Time),
+	}
+}
+
+func TestGetDockerStatsChecksDockerVersionAfterContainerList(t *testing.T) {
+	tests := []struct {
+		name            string
+		containerServer string
+		versionServer   string
+		versionBody     string
+		expectedGood    bool
+		expectedPodman  bool
+	}{
+		{
+			name:           "200 with good version on first try",
+			versionBody:    `{"Version":"25.0.1"}`,
+			expectedGood:   true,
+			expectedPodman: false,
+		},
+		{
+			name:           "200 with old version on first try",
+			versionBody:    `{"Version":"24.0.7"}`,
+			expectedGood:   false,
+			expectedPodman: false,
+		},
+		{
+			name:            "podman detected from server header",
+			containerServer: "Libpod/5.5.0",
+			expectedGood:    true,
+			expectedPodman:  true,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			requestCounts := map[string]int{}
+			server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				requestCounts[r.URL.EscapedPath()]++
+				switch r.URL.EscapedPath() {
+				case "/containers/json":
+					if tt.containerServer != "" {
+						w.Header().Set("Server", tt.containerServer)
+					}
+					w.WriteHeader(http.StatusOK)
+					fmt.Fprint(w, `[]`)
+				case "/version":
+					if tt.versionServer != "" {
+						w.Header().Set("Server", tt.versionServer)
+					}
+					w.WriteHeader(http.StatusOK)
+					fmt.Fprint(w, tt.versionBody)
+				default:
+					t.Fatalf("unexpected path: %s", r.URL.EscapedPath())
+				}
+			}))
+			defer server.Close()
+
+			dm := newDockerManagerForVersionTest(server)
+
+			stats, err := dm.getDockerStats(defaultCacheTimeMs)
+			require.NoError(t, err)
+			assert.Empty(t, stats)
+			assert.True(t, dm.dockerVersionChecked)
+			assert.Equal(t, tt.expectedGood, dm.goodDockerVersion)
+			assert.Equal(t, tt.expectedPodman, dm.usingPodman)
+			assert.Equal(t, 1, requestCounts["/containers/json"])
+			if tt.expectedPodman {
+				assert.Equal(t, 0, requestCounts["/version"])
+			} else {
+				assert.Equal(t, 1, requestCounts["/version"])
+			}
+
+			stats, err = dm.getDockerStats(defaultCacheTimeMs)
+			require.NoError(t, err)
+			assert.Empty(t, stats)
+			assert.Equal(t, tt.expectedGood, dm.goodDockerVersion)
+			assert.Equal(t, tt.expectedPodman, dm.usingPodman)
+			assert.Equal(t, 2, requestCounts["/containers/json"])
+			if tt.expectedPodman {
+				assert.Equal(t, 0, requestCounts["/version"])
+			} else {
+				assert.Equal(t, 1, requestCounts["/version"])
+			}
+		})
+	}
+
+}
+
+func TestGetDockerStatsRetriesVersionCheckUntilSuccess(t *testing.T) {
+	requestCounts := map[string]int{}
+	versionStatuses := []int{http.StatusServiceUnavailable, http.StatusOK}
+	versionBodies := []string{`"not ready"`, `{"Version":"25.1.0"}`}
+
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		requestCounts[r.URL.EscapedPath()]++
+		switch r.URL.EscapedPath() {
+		case "/containers/json":
+			w.WriteHeader(http.StatusOK)
+			fmt.Fprint(w, `[]`)
+		case "/version":
+			idx := requestCounts["/version"] - 1
+			if idx >= len(versionStatuses) {
+				idx = len(versionStatuses) - 1
+			}
+			w.WriteHeader(versionStatuses[idx])
+			fmt.Fprint(w, versionBodies[idx])
+		default:
+			t.Fatalf("unexpected path: %s", r.URL.EscapedPath())
+		}
+	}))
+	defer server.Close()
+
+	dm := newDockerManagerForVersionTest(server)
+
+	stats, err := dm.getDockerStats(defaultCacheTimeMs)
+	require.NoError(t, err)
+	assert.Empty(t, stats)
+	assert.False(t, dm.dockerVersionChecked)
+	assert.False(t, dm.goodDockerVersion)
+	assert.Equal(t, 1, requestCounts["/version"])
+
+	stats, err = dm.getDockerStats(defaultCacheTimeMs)
+	require.NoError(t, err)
+	assert.Empty(t, stats)
+	assert.True(t, dm.dockerVersionChecked)
+	assert.True(t, dm.goodDockerVersion)
+	assert.Equal(t, 2, requestCounts["/containers/json"])
+	assert.Equal(t, 2, requestCounts["/version"])
+
+	stats, err = dm.getDockerStats(defaultCacheTimeMs)
+	require.NoError(t, err)
+	assert.Empty(t, stats)
+	assert.Equal(t, 3, requestCounts["/containers/json"])
+	assert.Equal(t, 2, requestCounts["/version"])
 }
 
 func TestCycleCpuDeltas(t *testing.T) {
@@ -450,6 +875,7 @@ func TestDockerStatsWithMockData(t *testing.T) {
 		lastCpuReadTime:     make(map[uint16]map[string]time.Time),
 		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
 		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		lastNetworkReadTime: make(map[uint16]map[string]time.Time),
 		containerStatsMap:   make(map[string]*container.Stats),
 	}
 
@@ -595,23 +1021,22 @@ func TestNetworkStatsCalculationWithRealData(t *testing.T) {
 	dm := &dockerManager{
 		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
 		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		lastNetworkReadTime: make(map[uint16]map[string]time.Time),
 	}
 
 	ctr := &container.ApiInfo{IdShort: "test-container"}
 	cacheTimeMs := uint16(30000) // Test with 30 second cache
 
-	// Use exact timing for deterministic results
-	exactly1000msAgo := time.Now().Add(-1000 * time.Millisecond)
-	stats := &container.Stats{
-		PrevReadTime: exactly1000msAgo,
-	}
-
-	// First call sets baseline
-	sent1, recv1 := dm.calculateNetworkStats(ctr, apiStats1, stats, true, "test", cacheTimeMs)
+	// First call sets baseline (no previous read time, so rates should be 0)
+	sent1, recv1 := dm.calculateNetworkStats(ctr, apiStats1, "test", cacheTimeMs)
 	assert.Equal(t, uint64(0), sent1)
 	assert.Equal(t, uint64(0), recv1)
 
-	// Cycle to establish baseline for this cache time
+	// Record read time and cycle to establish baseline for this cache time
+	exactly1000msAgo := time.Now().Add(-1000 * time.Millisecond)
+	dm.lastNetworkReadTime[cacheTimeMs] = map[string]time.Time{
+		"test-container": exactly1000msAgo,
+	}
 	dm.cycleNetworkDeltasForCacheTime(cacheTimeMs)
 
 	// Calculate expected results precisely
@@ -622,7 +1047,7 @@ func TestNetworkStatsCalculationWithRealData(t *testing.T) {
 	expectedRecvRate := deltaRecv * 1000 / expectedElapsedMs // Should be exactly 1000000
 
 	// Second call with changed data
-	sent2, recv2 := dm.calculateNetworkStats(ctr, apiStats2, stats, true, "test", cacheTimeMs)
+	sent2, recv2 := dm.calculateNetworkStats(ctr, apiStats2, "test", cacheTimeMs)
 
 	// Should be exactly the expected rates (no tolerance needed)
 	assert.Equal(t, expectedSentRate, sent2)
@@ -630,12 +1055,13 @@ func TestNetworkStatsCalculationWithRealData(t *testing.T) {
 
 	// Bad speed cap: set absurd delta over 1ms and expect 0 due to cap
 	dm.cycleNetworkDeltasForCacheTime(cacheTimeMs)
-	stats.PrevReadTime = time.Now().Add(-1 * time.Millisecond)
+	dm.lastNetworkReadTime[cacheTimeMs]["test-container"] = time.Now().Add(-1 * time.Millisecond)
 	apiStats1.Networks["eth0"] = container.NetworkStats{TxBytes: 0, RxBytes: 0}
 	apiStats2.Networks["eth0"] = container.NetworkStats{TxBytes: 10 * 1024 * 1024 * 1024, RxBytes: 0} // 10GB delta
-	_, _ = dm.calculateNetworkStats(ctr, apiStats1, stats, true, "test", cacheTimeMs)                 // baseline
+	_, _ = dm.calculateNetworkStats(ctr, apiStats1, "test", cacheTimeMs)                              // baseline
 	dm.cycleNetworkDeltasForCacheTime(cacheTimeMs)
-	sent3, recv3 := dm.calculateNetworkStats(ctr, apiStats2, stats, true, "test", cacheTimeMs)
+	dm.lastNetworkReadTime[cacheTimeMs]["test-container"] = time.Now().Add(-1 * time.Millisecond)
+	sent3, recv3 := dm.calculateNetworkStats(ctr, apiStats2, "test", cacheTimeMs)
 	assert.Equal(t, uint64(0), sent3)
 	assert.Equal(t, uint64(0), recv3)
 }
@@ -656,6 +1082,7 @@ func TestContainerStatsEndToEndWithRealData(t *testing.T) {
 		lastCpuReadTime:     make(map[uint16]map[string]time.Time),
 		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
 		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		lastNetworkReadTime: make(map[uint16]map[string]time.Time),
 		containerStatsMap:   make(map[string]*container.Stats),
 	}
 
@@ -691,14 +1118,50 @@ func TestContainerStatsEndToEndWithRealData(t *testing.T) {
 	updateContainerStatsValues(testStats, cpuPct, usedMemory, 1000000, 500000, testTime)
 
 	assert.Equal(t, cpuPct, testStats.Cpu)
-	assert.Equal(t, bytesToMegabytes(float64(usedMemory)), testStats.Mem)
+	assert.Equal(t, utils.BytesToMegabytes(float64(usedMemory)), testStats.Mem)
 	assert.Equal(t, [2]uint64{1000000, 500000}, testStats.Bandwidth)
 	// Deprecated fields still populated for backward compatibility with older hubs
-	assert.Equal(t, bytesToMegabytes(1000000), testStats.NetworkSent)
-	assert.Equal(t, bytesToMegabytes(500000), testStats.NetworkRecv)
+	assert.Equal(t, utils.BytesToMegabytes(1000000), testStats.NetworkSent)
+	assert.Equal(t, utils.BytesToMegabytes(500000), testStats.NetworkRecv)
 	assert.Equal(t, testTime, testStats.PrevReadTime)
 }
 
+func TestGetLogsDetectsMultiplexedWithoutContentType(t *testing.T) {
+	// Docker multiplexed frame: [stream][0,0,0][len(4 bytes BE)][payload]
+	frame := []byte{
+		0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x05,
+		'H', 'e', 'l', 'l', 'o',
+	}
+	rt := &recordingRoundTripper{
+		statusCode: 200,
+		body:       string(frame),
+		// Intentionally omit content type to simulate Podman behavior.
+	}
+	dm := &dockerManager{
+		client: &http.Client{Transport: rt},
+	}
+
+	logs, err := dm.getLogs(context.Background(), "abcdef123456")
+	require.NoError(t, err)
+	assert.Equal(t, "Hello", logs)
+}
+
+func TestGetLogsDoesNotMisclassifyRawStreamAsMultiplexed(t *testing.T) {
+	// Starts with 0x01, but doesn't match Docker frame signature (reserved bytes aren't all zero).
+	raw := []byte{0x01, 0x02, 0x03, 0x04, 'r', 'a', 'w'}
+	rt := &recordingRoundTripper{
+		statusCode: 200,
+		body:       string(raw),
+	}
+	dm := &dockerManager{
+		client: &http.Client{Transport: rt},
+	}
+
+	logs, err := dm.getLogs(context.Background(), "abcdef123456")
+	require.NoError(t, err)
+	assert.Equal(t, raw, []byte(logs))
+}
+
 func TestEdgeCasesWithRealData(t *testing.T) {
 	// Test with minimal container stats
 	minimalStats := &container.ApiStats{
@@ -741,6 +1204,7 @@ func TestDockerStatsWorkflow(t *testing.T) {
 		lastCpuSystem:       make(map[uint16]map[string]uint64),
 		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
 		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		lastNetworkReadTime: make(map[uint16]map[string]time.Time),
 		containerStatsMap:   make(map[string]*container.Stats),
 	}
 
@@ -920,6 +1384,18 @@ func TestParseDockerStatus(t *testing.T) {
 			expectedStatus: "",
 			expectedHealth: container.DockerHealthNone,
 		},
+		{
+			name:           "status health with health: prefix",
+			input:          "Up 5 minutes (health: starting)",
+			expectedStatus: "Up 5 minutes",
+			expectedHealth: container.DockerHealthStarting,
+		},
+		{
+			name:           "status health with health status: prefix",
+			input:          "Up 10 minutes (health status: unhealthy)",
+			expectedStatus: "Up 10 minutes",
+			expectedHealth: container.DockerHealthUnhealthy,
+		},
 	}
 
 	for _, tt := range tests {
@@ -931,6 +1407,85 @@ func TestParseDockerStatus(t *testing.T) {
 	}
 }
 
+func TestParseDockerHealthStatus(t *testing.T) {
+	tests := []struct {
+		input          string
+		expectedHealth container.DockerHealth
+		expectedOk     bool
+	}{
+		{"healthy", container.DockerHealthHealthy, true},
+		{"unhealthy", container.DockerHealthUnhealthy, true},
+		{"starting", container.DockerHealthStarting, true},
+		{"none", container.DockerHealthNone, true},
+		{" Healthy ", container.DockerHealthHealthy, true},
+		{"unknown", container.DockerHealthNone, false},
+		{"", container.DockerHealthNone, false},
+	}
+	for _, tt := range tests {
+		t.Run(tt.input, func(t *testing.T) {
+			health, ok := parseDockerHealthStatus(tt.input)
+			assert.Equal(t, tt.expectedHealth, health)
+			assert.Equal(t, tt.expectedOk, ok)
+		})
+	}
+}
+
+func TestUpdateContainerStatsUsesPodmanInspectHealthFallback(t *testing.T) {
+	var requestedPaths []string
+	dm := &dockerManager{
+		client: &http.Client{Transport: roundTripFunc(func(req *http.Request) (*http.Response, error) {
+			requestedPaths = append(requestedPaths, req.URL.EscapedPath())
+			switch req.URL.EscapedPath() {
+			case "/containers/0123456789ab/stats":
+				return &http.Response{
+					StatusCode: http.StatusOK,
+					Status:     "200 OK",
+					Header:     make(http.Header),
+					Body: io.NopCloser(strings.NewReader(`{
+						"read":"2026-03-15T21:26:59Z",
+						"cpu_stats":{"cpu_usage":{"total_usage":1000},"system_cpu_usage":2000},
+						"memory_stats":{"usage":1048576,"stats":{"inactive_file":262144}},
+						"networks":{"eth0":{"rx_bytes":0,"tx_bytes":0}}
+					}`)),
+					Request: req,
+				}, nil
+			case "/containers/0123456789ab/json":
+				return &http.Response{
+					StatusCode: http.StatusOK,
+					Status:     "200 OK",
+					Header:     make(http.Header),
+					Body:       io.NopCloser(strings.NewReader(`{"State":{"Health":{"Status":"healthy"}}}`)),
+					Request:    req,
+				}, nil
+			default:
+				return nil, fmt.Errorf("unexpected path: %s", req.URL.EscapedPath())
+			}
+		})},
+		containerStatsMap:   make(map[string]*container.Stats),
+		apiStats:            &container.ApiStats{},
+		usingPodman:         true,
+		lastCpuContainer:    make(map[uint16]map[string]uint64),
+		lastCpuSystem:       make(map[uint16]map[string]uint64),
+		lastCpuReadTime:     make(map[uint16]map[string]time.Time),
+		networkSentTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		networkRecvTrackers: make(map[uint16]*deltatracker.DeltaTracker[string, uint64]),
+		lastNetworkReadTime: make(map[uint16]map[string]time.Time),
+	}
+
+	ctr := &container.ApiInfo{
+		IdShort: "0123456789ab",
+		Names:   []string{"/beszel"},
+		Status:  "Up 2 minutes",
+		Image:   "beszel:latest",
+	}
+
+	err := dm.updateContainerStats(ctr, defaultCacheTimeMs)
+	require.NoError(t, err)
+	assert.Equal(t, []string{"/containers/0123456789ab/stats", "/containers/0123456789ab/json"}, requestedPaths)
+	assert.Equal(t, container.DockerHealthHealthy, dm.containerStatsMap[ctr.IdShort].Health)
+	assert.Equal(t, "Up 2 minutes", dm.containerStatsMap[ctr.IdShort].Status)
+}
+
 func TestConstantsAndUtilityFunctions(t *testing.T) {
 	// Test constants are properly defined
 	assert.Equal(t, uint16(60000), defaultCacheTimeMs)
@@ -940,13 +1495,13 @@ func TestConstantsAndUtilityFunctions(t *testing.T) {
 	assert.Equal(t, 5*1024*1024, maxTotalLogSize)               // 5MB
 
 	// Test utility functions
-	assert.Equal(t, 1.5, twoDecimals(1.499))
-	assert.Equal(t, 1.5, twoDecimals(1.5))
-	assert.Equal(t, 1.5, twoDecimals(1.501))
+	assert.Equal(t, 1.5, utils.TwoDecimals(1.499))
+	assert.Equal(t, 1.5, utils.TwoDecimals(1.5))
+	assert.Equal(t, 1.5, utils.TwoDecimals(1.501))
 
-	assert.Equal(t, 1.0, bytesToMegabytes(1048576)) // 1 MB
-	assert.Equal(t, 0.5, bytesToMegabytes(524288))  // 512 KB
-	assert.Equal(t, 0.0, bytesToMegabytes(0))
+	assert.Equal(t, 1.0, utils.BytesToMegabytes(1048576)) // 1 MB
+	assert.Equal(t, 0.5, utils.BytesToMegabytes(524288))  // 512 KB
+	assert.Equal(t, 0.0, utils.BytesToMegabytes(0))
 }
 
 func TestDecodeDockerLogStream(t *testing.T) {
@@ -1246,3 +1801,99 @@ func TestAnsiEscapePattern(t *testing.T) {
 		})
 	}
 }
+
+func TestConvertContainerPortsToString(t *testing.T) {
+	type port = struct {
+		PublicPort uint16
+		IP         string
+	}
+	tests := []struct {
+		name     string
+		ports    []port
+		expected string
+	}{
+		{
+			name:     "empty ports",
+			ports:    nil,
+			expected: "",
+		},
+		{
+			name: "single port",
+			ports: []port{
+				{PublicPort: 80, IP: "0.0.0.0"},
+			},
+			expected: "80",
+		},
+		{
+			name: "single port with non-default IP",
+			ports: []port{
+				{PublicPort: 80, IP: "1.2.3.4"},
+			},
+			expected: "1.2.3.4:80",
+		},
+		{
+			name: "ipv6 default ip",
+			ports: []port{
+				{PublicPort: 80, IP: "::"},
+			},
+			expected: "80",
+		},
+		{
+			name: "zero PublicPort is skipped",
+			ports: []port{
+				{PublicPort: 0, IP: "0.0.0.0"},
+				{PublicPort: 80, IP: "0.0.0.0"},
+			},
+			expected: "80",
+		},
+		{
+			name: "ports sorted ascending by PublicPort",
+			ports: []port{
+				{PublicPort: 443, IP: "0.0.0.0"},
+				{PublicPort: 80, IP: "0.0.0.0"},
+				{PublicPort: 8080, IP: "0.0.0.0"},
+			},
+			expected: "80, 443, 8080",
+		},
+		{
+			name: "duplicates are deduplicated",
+			ports: []port{
+				{PublicPort: 80, IP: "0.0.0.0"},
+				{PublicPort: 80, IP: "0.0.0.0"},
+				{PublicPort: 443, IP: "0.0.0.0"},
+			},
+			expected: "80, 443",
+		},
+		{
+			name: "multiple ports with different IPs",
+			ports: []port{
+				{PublicPort: 80, IP: "0.0.0.0"},
+				{PublicPort: 443, IP: "1.2.3.4"},
+			},
+			expected: "80, 1.2.3.4:443",
+		},
+		{
+			name: "ports slice is nilled after call",
+			ports: []port{
+				{PublicPort: 8080, IP: "0.0.0.0"},
+			},
+			expected: "8080",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			ctr := &container.ApiInfo{}
+			for _, p := range tt.ports {
+				ctr.Ports = append(ctr.Ports, struct {
+					PublicPort uint16
+					IP         string
+				}{PublicPort: p.PublicPort, IP: p.IP})
+			}
+			result := convertContainerPortsToString(ctr)
+			assert.Equal(t, tt.expected, result)
+			// Ports slice must be cleared to prevent bleed-over into the next response
+			assert.Nil(t, ctr.Ports, "ctr.Ports should be nil after formatContainerPorts")
+		})
+	}
+}

agent/emmc_common.go

@@ -0,0 +1,95 @@
+package agent
+
+import (
+	"fmt"
+	"strconv"
+	"strings"
+)
+
+func isEmmcBlockName(name string) bool {
+	if !strings.HasPrefix(name, "mmcblk") {
+		return false
+	}
+	suffix := strings.TrimPrefix(name, "mmcblk")
+	if suffix == "" {
+		return false
+	}
+	for _, c := range suffix {
+		if c < '0' || c > '9' {
+			return false
+		}
+	}
+	return true
+}
+
+func parseHexOrDecByte(s string) (uint8, bool) {
+	s = strings.TrimSpace(s)
+	if s == "" {
+		return 0, false
+	}
+	base := 10
+	if strings.HasPrefix(s, "0x") || strings.HasPrefix(s, "0X") {
+		base = 16
+		s = s[2:]
+	}
+	parsed, err := strconv.ParseUint(s, base, 8)
+	if err != nil {
+		return 0, false
+	}
+	return uint8(parsed), true
+}
+
+func parseHexBytePair(s string) (uint8, uint8, bool) {
+	fields := strings.Fields(s)
+	if len(fields) < 2 {
+		return 0, 0, false
+	}
+	a, okA := parseHexOrDecByte(fields[0])
+	b, okB := parseHexOrDecByte(fields[1])
+	if !okA && !okB {
+		return 0, 0, false
+	}
+	return a, b, true
+}
+
+func emmcSmartStatus(preEOL uint8) string {
+	switch preEOL {
+	case 0x01:
+		return "PASSED"
+	case 0x02:
+		return "WARNING"
+	case 0x03:
+		return "FAILED"
+	default:
+		return "UNKNOWN"
+	}
+}
+
+func emmcPreEOLString(preEOL uint8) string {
+	switch preEOL {
+	case 0x01:
+		return "0x01 (normal)"
+	case 0x02:
+		return "0x02 (warning)"
+	case 0x03:
+		return "0x03 (urgent)"
+	default:
+		return fmt.Sprintf("0x%02x", preEOL)
+	}
+}
+
+func emmcLifeTimeString(v uint8) string {
+	// JEDEC eMMC: 0x01..0x0A => 0-100% used in 10% steps, 0x0B => exceeded.
+	switch {
+	case v == 0:
+		return "0x00 (not reported)"
+	case v >= 0x01 && v <= 0x0A:
+		low := int(v-1) * 10
+		high := int(v) * 10
+		return fmt.Sprintf("0x%02x (%d-%d%% used)", v, low, high)
+	case v == 0x0B:
+		return "0x0b (>100% used)"
+	default:
+		return fmt.Sprintf("0x%02x", v)
+	}
+}

agent/emmc_common_test.go

@@ -0,0 +1,78 @@
+package agent
+
+import "testing"
+
+func TestParseHexOrDecByte(t *testing.T) {
+	tests := []struct {
+		in   string
+		want uint8
+		ok   bool
+	}{
+		{"0x01", 1, true},
+		{"0X0b", 11, true},
+		{"01", 1, true},
+		{" 3 ", 3, true},
+		{"", 0, false},
+		{"0x", 0, false},
+		{"nope", 0, false},
+	}
+
+	for _, tt := range tests {
+		got, ok := parseHexOrDecByte(tt.in)
+		if ok != tt.ok || got != tt.want {
+			t.Fatalf("parseHexOrDecByte(%q) = (%d,%v), want (%d,%v)", tt.in, got, ok, tt.want, tt.ok)
+		}
+	}
+}
+
+func TestParseHexBytePair(t *testing.T) {
+	a, b, ok := parseHexBytePair("0x01 0x02\n")
+	if !ok || a != 1 || b != 2 {
+		t.Fatalf("parseHexBytePair hex = (%d,%d,%v), want (1,2,true)", a, b, ok)
+	}
+
+	a, b, ok = parseHexBytePair("01 02")
+	if !ok || a != 1 || b != 2 {
+		t.Fatalf("parseHexBytePair dec = (%d,%d,%v), want (1,2,true)", a, b, ok)
+	}
+
+	_, _, ok = parseHexBytePair("0x01")
+	if ok {
+		t.Fatalf("parseHexBytePair short input ok=true, want false")
+	}
+}
+
+func TestEmmcSmartStatus(t *testing.T) {
+	if got := emmcSmartStatus(0x01); got != "PASSED" {
+		t.Fatalf("emmcSmartStatus(0x01) = %q, want PASSED", got)
+	}
+	if got := emmcSmartStatus(0x02); got != "WARNING" {
+		t.Fatalf("emmcSmartStatus(0x02) = %q, want WARNING", got)
+	}
+	if got := emmcSmartStatus(0x03); got != "FAILED" {
+		t.Fatalf("emmcSmartStatus(0x03) = %q, want FAILED", got)
+	}
+	if got := emmcSmartStatus(0x00); got != "UNKNOWN" {
+		t.Fatalf("emmcSmartStatus(0x00) = %q, want UNKNOWN", got)
+	}
+}
+
+func TestIsEmmcBlockName(t *testing.T) {
+	cases := []struct {
+		name string
+		ok   bool
+	}{
+		{"mmcblk0", true},
+		{"mmcblk1", true},
+		{"mmcblk10", true},
+		{"mmcblk0p1", false},
+		{"sda", false},
+		{"mmcblk", false},
+		{"mmcblkA", false},
+	}
+	for _, c := range cases {
+		if got := isEmmcBlockName(c.name); got != c.ok {
+			t.Fatalf("isEmmcBlockName(%q) = %v, want %v", c.name, got, c.ok)
+		}
+	}
+}

agent/emmc_linux.go

@@ -0,0 +1,215 @@
+//go:build linux
+
+package agent
+
+import (
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+
+	"github.com/henrygd/beszel/agent/utils"
+	"github.com/henrygd/beszel/internal/entities/smart"
+)
+
+// emmcSysfsRoot is a test hook; production value is "/sys".
+var emmcSysfsRoot = "/sys"
+
+type emmcHealth struct {
+	model    string
+	serial   string
+	revision string
+	capacity uint64
+	preEOL   uint8
+	lifeA    uint8
+	lifeB    uint8
+}
+
+func scanEmmcDevices() []*DeviceInfo {
+	blockDir := filepath.Join(emmcSysfsRoot, "class", "block")
+	entries, err := os.ReadDir(blockDir)
+	if err != nil {
+		return nil
+	}
+
+	devices := make([]*DeviceInfo, 0, 2)
+	for _, ent := range entries {
+		name := ent.Name()
+		if !isEmmcBlockName(name) {
+			continue
+		}
+
+		deviceDir := filepath.Join(blockDir, name, "device")
+		if !hasEmmcHealthFiles(deviceDir) {
+			continue
+		}
+
+		devPath := filepath.Join("/dev", name)
+		devices = append(devices, &DeviceInfo{
+			Name:     devPath,
+			Type:     "emmc",
+			InfoName: devPath + " [eMMC]",
+			Protocol: "MMC",
+		})
+	}
+
+	return devices
+}
+
+func (sm *SmartManager) collectEmmcHealth(deviceInfo *DeviceInfo) (bool, error) {
+	if deviceInfo == nil || deviceInfo.Name == "" {
+		return false, nil
+	}
+
+	base := filepath.Base(deviceInfo.Name)
+	if !isEmmcBlockName(base) && !strings.EqualFold(deviceInfo.Type, "emmc") && !strings.EqualFold(deviceInfo.Type, "mmc") {
+		return false, nil
+	}
+
+	health, ok := readEmmcHealth(base)
+	if !ok {
+		return false, nil
+	}
+
+	// Normalize the device type to keep pruning logic stable across refreshes.
+	deviceInfo.Type = "emmc"
+
+	key := health.serial
+	if key == "" {
+		key = filepath.Join("/dev", base)
+	}
+
+	status := emmcSmartStatus(health.preEOL)
+
+	attrs := []*smart.SmartAttribute{
+		{
+			Name:      "PreEOLInfo",
+			RawValue:  uint64(health.preEOL),
+			RawString: emmcPreEOLString(health.preEOL),
+		},
+		{
+			Name:      "DeviceLifeTimeEstA",
+			RawValue:  uint64(health.lifeA),
+			RawString: emmcLifeTimeString(health.lifeA),
+		},
+		{
+			Name:      "DeviceLifeTimeEstB",
+			RawValue:  uint64(health.lifeB),
+			RawString: emmcLifeTimeString(health.lifeB),
+		},
+	}
+
+	sm.Lock()
+	defer sm.Unlock()
+
+	if _, exists := sm.SmartDataMap[key]; !exists {
+		sm.SmartDataMap[key] = &smart.SmartData{}
+	}
+
+	data := sm.SmartDataMap[key]
+	data.ModelName = health.model
+	data.SerialNumber = health.serial
+	data.FirmwareVersion = health.revision
+	data.Capacity = health.capacity
+	data.Temperature = 0
+	data.SmartStatus = status
+	data.DiskName = filepath.Join("/dev", base)
+	data.DiskType = "emmc"
+	data.Attributes = attrs
+
+	return true, nil
+}
+
+func readEmmcHealth(blockName string) (emmcHealth, bool) {
+	var out emmcHealth
+
+	if !isEmmcBlockName(blockName) {
+		return out, false
+	}
+
+	deviceDir := filepath.Join(emmcSysfsRoot, "class", "block", blockName, "device")
+	preEOL, okPre := readHexByteFile(filepath.Join(deviceDir, "pre_eol_info"))
+
+	// Some kernels expose EXT_CSD lifetime via "life_time" (two bytes), others as
+	// separate files. Support both.
+	lifeA, lifeB, okLife := readLifeTime(deviceDir)
+
+	if !okPre && !okLife {
+		return out, false
+	}
+
+	out.preEOL = preEOL
+	out.lifeA = lifeA
+	out.lifeB = lifeB
+
+	out.model = utils.ReadStringFile(filepath.Join(deviceDir, "name"))
+	out.serial = utils.ReadStringFile(filepath.Join(deviceDir, "serial"))
+	out.revision = utils.ReadStringFile(filepath.Join(deviceDir, "prv"))
+
+	if capBytes, ok := readBlockCapacityBytes(blockName); ok {
+		out.capacity = capBytes
+	}
+
+	return out, true
+}
+
+func readLifeTime(deviceDir string) (uint8, uint8, bool) {
+	if content, ok := utils.ReadStringFileOK(filepath.Join(deviceDir, "life_time")); ok {
+		a, b, ok := parseHexBytePair(content)
+		return a, b, ok
+	}
+
+	a, okA := readHexByteFile(filepath.Join(deviceDir, "device_life_time_est_typ_a"))
+	b, okB := readHexByteFile(filepath.Join(deviceDir, "device_life_time_est_typ_b"))
+	if okA || okB {
+		return a, b, true
+	}
+	return 0, 0, false
+}
+
+func readBlockCapacityBytes(blockName string) (uint64, bool) {
+	sizePath := filepath.Join(emmcSysfsRoot, "class", "block", blockName, "size")
+	lbsPath := filepath.Join(emmcSysfsRoot, "class", "block", blockName, "queue", "logical_block_size")
+
+	sizeStr, ok := utils.ReadStringFileOK(sizePath)
+	if !ok {
+		return 0, false
+	}
+	sectors, err := strconv.ParseUint(sizeStr, 10, 64)
+	if err != nil || sectors == 0 {
+		return 0, false
+	}
+
+	lbsStr, ok := utils.ReadStringFileOK(lbsPath)
+	logicalBlockSize := uint64(512)
+	if ok {
+		if parsed, err := strconv.ParseUint(lbsStr, 10, 64); err == nil && parsed > 0 {
+			logicalBlockSize = parsed
+		}
+	}
+
+	return sectors * logicalBlockSize, true
+}
+
+func readHexByteFile(path string) (uint8, bool) {
+	content, ok := utils.ReadStringFileOK(path)
+	if !ok {
+		return 0, false
+	}
+	b, ok := parseHexOrDecByte(content)
+	return b, ok
+}
+
+func hasEmmcHealthFiles(deviceDir string) bool {
+	entries, err := os.ReadDir(deviceDir)
+	if err != nil {
+		return false
+	}
+	for _, ent := range entries {
+		switch ent.Name() {
+		case "pre_eol_info", "life_time", "device_life_time_est_typ_a", "device_life_time_est_typ_b":
+			return true
+		}
+	}
+	return false
+}

agent/emmc_linux_test.go

@@ -0,0 +1,80 @@
+//go:build linux
+
+package agent
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/henrygd/beszel/internal/entities/smart"
+)
+
+func TestEmmcMockSysfsScanAndCollect(t *testing.T) {
+	tmp := t.TempDir()
+	prev := emmcSysfsRoot
+	emmcSysfsRoot = tmp
+	t.Cleanup(func() { emmcSysfsRoot = prev })
+
+	// Fake: /sys/class/block/mmcblk0
+	mmcDeviceDir := filepath.Join(tmp, "class", "block", "mmcblk0", "device")
+	mmcQueueDir := filepath.Join(tmp, "class", "block", "mmcblk0", "queue")
+	if err := os.MkdirAll(mmcDeviceDir, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.MkdirAll(mmcQueueDir, 0o755); err != nil {
+		t.Fatal(err)
+	}
+
+	write := func(path, content string) {
+		t.Helper()
+		if err := os.WriteFile(path, []byte(content), 0o644); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	write(filepath.Join(mmcDeviceDir, "pre_eol_info"), "0x02\n")
+	write(filepath.Join(mmcDeviceDir, "life_time"), "0x04 0x05\n")
+	write(filepath.Join(mmcDeviceDir, "name"), "H26M52103FMR\n")
+	write(filepath.Join(mmcDeviceDir, "serial"), "01234567\n")
+	write(filepath.Join(mmcDeviceDir, "prv"), "0x08\n")
+	write(filepath.Join(mmcQueueDir, "logical_block_size"), "512\n")
+	write(filepath.Join(tmp, "class", "block", "mmcblk0", "size"), "1024\n") // sectors
+
+	devs := scanEmmcDevices()
+	if len(devs) != 1 {
+		t.Fatalf("scanEmmcDevices() = %d devices, want 1", len(devs))
+	}
+	if devs[0].Name != "/dev/mmcblk0" || devs[0].Type != "emmc" {
+		t.Fatalf("scanEmmcDevices()[0] = %+v, want Name=/dev/mmcblk0 Type=emmc", devs[0])
+	}
+
+	sm := &SmartManager{SmartDataMap: map[string]*smart.SmartData{}}
+	ok, err := sm.collectEmmcHealth(devs[0])
+	if err != nil || !ok {
+		t.Fatalf("collectEmmcHealth() = (ok=%v, err=%v), want (true,nil)", ok, err)
+	}
+	if len(sm.SmartDataMap) != 1 {
+		t.Fatalf("SmartDataMap len=%d, want 1", len(sm.SmartDataMap))
+	}
+	var got *smart.SmartData
+	for _, v := range sm.SmartDataMap {
+		got = v
+		break
+	}
+	if got == nil {
+		t.Fatalf("SmartDataMap value nil")
+	}
+	if got.DiskType != "emmc" || got.DiskName != "/dev/mmcblk0" {
+		t.Fatalf("disk fields = (type=%q name=%q), want (emmc,/dev/mmcblk0)", got.DiskType, got.DiskName)
+	}
+	if got.SmartStatus != "WARNING" {
+		t.Fatalf("SmartStatus=%q, want WARNING", got.SmartStatus)
+	}
+	if got.SerialNumber != "01234567" || got.ModelName == "" || got.Capacity == 0 {
+		t.Fatalf("identity fields = (model=%q serial=%q cap=%d), want non-empty model, serial 01234567, cap>0", got.ModelName, got.SerialNumber, got.Capacity)
+	}
+	if len(got.Attributes) < 3 {
+		t.Fatalf("attributes len=%d, want >= 3", len(got.Attributes))
+	}
+}

agent/emmc_stub.go

@@ -0,0 +1,14 @@
+//go:build !linux
+
+package agent
+
+// Non-Linux builds: eMMC health via sysfs is not available.
+
+func scanEmmcDevices() []*DeviceInfo {
+	return nil
+}
+
+func (sm *SmartManager) collectEmmcHealth(deviceInfo *DeviceInfo) (bool, error) {
+	return false, nil
+}
+

agent/fingerprint.go

@@ -0,0 +1,87 @@
+package agent
+
+import (
+	"crypto/sha256"
+	"encoding/hex"
+	"errors"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"github.com/shirou/gopsutil/v4/cpu"
+	"github.com/shirou/gopsutil/v4/host"
+)
+
+const fingerprintFileName = "fingerprint"
+
+// knownBadUUID is a commonly known "product_uuid" that is not unique across systems.
+const knownBadUUID = "03000200-0400-0500-0006-000700080009"
+
+// GetFingerprint returns the agent fingerprint. It first tries to read a saved
+// fingerprint from the data directory. If not found (or dataDir is empty), it
+// generates one from system properties. The hostname and cpuModel parameters are
+// used as fallback material if host.HostID() fails. If either is empty, they
+// are fetched from the system automatically.
+//
+// If a new fingerprint is generated and a dataDir is provided, it is saved.
+func GetFingerprint(dataDir, hostname, cpuModel string) string {
+	if dataDir != "" {
+		if fp, err := readFingerprint(dataDir); err == nil {
+			return fp
+		}
+	}
+	fp := generateFingerprint(hostname, cpuModel)
+	if dataDir != "" {
+		_ = SaveFingerprint(dataDir, fp)
+	}
+	return fp
+}
+
+// generateFingerprint creates a fingerprint from system properties.
+// It tries host.HostID() first, falling back to hostname + cpuModel.
+// If hostname or cpuModel are empty, they are fetched from the system.
+func generateFingerprint(hostname, cpuModel string) string {
+	fingerprint, err := host.HostID()
+	if err != nil || fingerprint == "" || fingerprint == knownBadUUID {
+		if hostname == "" {
+			hostname, _ = os.Hostname()
+		}
+		if cpuModel == "" {
+			if info, err := cpu.Info(); err == nil && len(info) > 0 {
+				cpuModel = info[0].ModelName
+			}
+		}
+		fingerprint = hostname + cpuModel
+	}
+
+	sum := sha256.Sum256([]byte(fingerprint))
+	return hex.EncodeToString(sum[:24])
+}
+
+// readFingerprint reads the saved fingerprint from the data directory.
+func readFingerprint(dataDir string) (string, error) {
+	fp, err := os.ReadFile(filepath.Join(dataDir, fingerprintFileName))
+	if err != nil {
+		return "", err
+	}
+	s := strings.TrimSpace(string(fp))
+	if s == "" {
+		return "", errors.New("fingerprint file is empty")
+	}
+	return s, nil
+}
+
+// SaveFingerprint writes the fingerprint to the data directory.
+func SaveFingerprint(dataDir, fingerprint string) error {
+	return os.WriteFile(filepath.Join(dataDir, fingerprintFileName), []byte(fingerprint), 0o644)
+}
+
+// DeleteFingerprint removes the saved fingerprint file from the data directory.
+// Returns nil if the file does not exist (idempotent).
+func DeleteFingerprint(dataDir string) error {
+	err := os.Remove(filepath.Join(dataDir, fingerprintFileName))
+	if errors.Is(err, os.ErrNotExist) {
+		return nil
+	}
+	return err
+}

agent/fingerprint_test.go

@@ -0,0 +1,102 @@
+//go:build testing
+
+package agent
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestGetFingerprint(t *testing.T) {
+	t.Run("reads existing fingerprint from file", func(t *testing.T) {
+		dir := t.TempDir()
+		expected := "abc123def456"
+		err := os.WriteFile(filepath.Join(dir, fingerprintFileName), []byte(expected), 0644)
+		require.NoError(t, err)
+
+		fp := GetFingerprint(dir, "", "")
+		assert.Equal(t, expected, fp)
+	})
+
+	t.Run("trims whitespace from file", func(t *testing.T) {
+		dir := t.TempDir()
+		err := os.WriteFile(filepath.Join(dir, fingerprintFileName), []byte("  abc123  \n"), 0644)
+		require.NoError(t, err)
+
+		fp := GetFingerprint(dir, "", "")
+		assert.Equal(t, "abc123", fp)
+	})
+
+	t.Run("generates fingerprint when file does not exist", func(t *testing.T) {
+		dir := t.TempDir()
+		fp := GetFingerprint(dir, "", "")
+		assert.NotEmpty(t, fp)
+	})
+
+	t.Run("generates fingerprint when dataDir is empty", func(t *testing.T) {
+		fp := GetFingerprint("", "", "")
+		assert.NotEmpty(t, fp)
+	})
+
+	t.Run("generates consistent fingerprint for same inputs", func(t *testing.T) {
+		fp1 := GetFingerprint("", "myhost", "mycpu")
+		fp2 := GetFingerprint("", "myhost", "mycpu")
+		assert.Equal(t, fp1, fp2)
+	})
+
+	t.Run("prefers saved fingerprint over generated", func(t *testing.T) {
+		dir := t.TempDir()
+		require.NoError(t, SaveFingerprint(dir, "saved-fp"))
+
+		fp := GetFingerprint(dir, "anyhost", "anycpu")
+		assert.Equal(t, "saved-fp", fp)
+	})
+}
+
+func TestSaveFingerprint(t *testing.T) {
+	t.Run("saves fingerprint to file", func(t *testing.T) {
+		dir := t.TempDir()
+		err := SaveFingerprint(dir, "abc123")
+		require.NoError(t, err)
+
+		content, err := os.ReadFile(filepath.Join(dir, fingerprintFileName))
+		require.NoError(t, err)
+		assert.Equal(t, "abc123", string(content))
+	})
+
+	t.Run("overwrites existing fingerprint", func(t *testing.T) {
+		dir := t.TempDir()
+		require.NoError(t, SaveFingerprint(dir, "old"))
+		require.NoError(t, SaveFingerprint(dir, "new"))
+
+		content, err := os.ReadFile(filepath.Join(dir, fingerprintFileName))
+		require.NoError(t, err)
+		assert.Equal(t, "new", string(content))
+	})
+}
+
+func TestDeleteFingerprint(t *testing.T) {
+	t.Run("deletes existing fingerprint", func(t *testing.T) {
+		dir := t.TempDir()
+		fp := filepath.Join(dir, fingerprintFileName)
+		err := os.WriteFile(fp, []byte("abc123"), 0644)
+		require.NoError(t, err)
+
+		err = DeleteFingerprint(dir)
+		require.NoError(t, err)
+
+		// Verify file is gone
+		_, err = os.Stat(fp)
+		assert.True(t, os.IsNotExist(err))
+	})
+
+	t.Run("no error when file does not exist", func(t *testing.T) {
+		dir := t.TempDir()
+		err := DeleteFingerprint(dir)
+		assert.NoError(t, err)
+	})
+}

agent/gpu.go

@@ -9,25 +9,26 @@ import (
 	"maps"
 	"os/exec"
 	"regexp"
+	"runtime"
 	"strconv"
 	"strings"
 	"sync"
 	"time"
 
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"
 )
 
 const (
 	// Commands
-	nvidiaSmiCmd  string = "nvidia-smi"
-	rocmSmiCmd    string = "rocm-smi"
-	amdgpuCmd     string = "amdgpu" // internal cmd for sysfs collection
-	tegraStatsCmd string = "tegrastats"
+	nvidiaSmiCmd    string = "nvidia-smi"
+	rocmSmiCmd      string = "rocm-smi"
+	tegraStatsCmd   string = "tegrastats"
+	nvtopCmd        string = "nvtop"
+	powermetricsCmd string = "powermetrics"
+	macmonCmd       string = "macmon"
+	noGPUFoundMsg   string = "no GPU found - see https://beszel.dev/guide/gpu"
 
-	// Polling intervals
-	nvidiaSmiInterval  string        = "4"    // in seconds
-	tegraStatsInterval string        = "3700" // in milliseconds
-	rocmSmiInterval    time.Duration = 4300 * time.Millisecond
 	// Command retry and timeout constants
 	retryWaitTime     time.Duration = 5 * time.Second
 	maxFailureRetries int           = 5
@@ -40,13 +41,7 @@ const (
 // GPUManager manages data collection for GPUs (either Nvidia or AMD)
 type GPUManager struct {
 	sync.Mutex
-	nvidiaSmi     bool
-	rocmSmi       bool
-	amdgpu        bool
-	tegrastats    bool
-	intelGpuStats bool
-	nvml          bool
-	GpuDataMap    map[string]*system.GPUData
+	GpuDataMap map[string]*system.GPUData
 	// lastAvgData stores the last calculated averages for each GPU
 	// Used when a collection happens before new data arrives (Count == 0)
 	lastAvgData map[string]system.GPUData
@@ -87,6 +82,58 @@ type gpuCollector struct {
 
 var errNoValidData = fmt.Errorf("no valid GPU data found") // Error for missing data
 
+// collectorSource identifies a selectable GPU collector in GPU_COLLECTOR.
+type collectorSource string
+
+const (
+	collectorSourceNVTop        collectorSource = collectorSource(nvtopCmd)
+	collectorSourceNVML         collectorSource = "nvml"
+	collectorSourceNvidiaSMI    collectorSource = collectorSource(nvidiaSmiCmd)
+	collectorSourceIntelGpuTop  collectorSource = collectorSource(intelGpuStatsCmd)
+	collectorSourceAmdSysfs     collectorSource = "amd_sysfs"
+	collectorSourceRocmSMI      collectorSource = collectorSource(rocmSmiCmd)
+	collectorSourceMacmon       collectorSource = collectorSource(macmonCmd)
+	collectorSourcePowermetrics collectorSource = collectorSource(powermetricsCmd)
+	collectorGroupNvidia        string          = "nvidia"
+	collectorGroupIntel         string          = "intel"
+	collectorGroupAmd           string          = "amd"
+	collectorGroupApple         string          = "apple"
+)
+
+func isValidCollectorSource(source collectorSource) bool {
+	switch source {
+	case collectorSourceNVTop,
+		collectorSourceNVML,
+		collectorSourceNvidiaSMI,
+		collectorSourceIntelGpuTop,
+		collectorSourceAmdSysfs,
+		collectorSourceRocmSMI,
+		collectorSourceMacmon,
+		collectorSourcePowermetrics:
+		return true
+	}
+	return false
+}
+
+// gpuCapabilities describes detected GPU tooling and sysfs support on the host.
+type gpuCapabilities struct {
+	hasNvidiaSmi    bool
+	hasRocmSmi      bool
+	hasAmdSysfs     bool
+	hasTegrastats   bool
+	hasIntelGpuTop  bool
+	hasNvtop        bool
+	hasMacmon       bool
+	hasPowermetrics bool
+}
+
+type collectorDefinition struct {
+	group              string
+	available          bool
+	start              func(onFailure func()) bool
+	deprecationWarning string
+}
+
 // starts and manages the ongoing collection of GPU data for the specified GPU management utility
 func (c *gpuCollector) start() {
 	for {
@@ -245,8 +292,8 @@ func (gm *GPUManager) parseAmdData(output []byte) bool {
 		}
 		gpu := gm.GpuDataMap[id]
 		gpu.Temperature, _ = strconv.ParseFloat(v.Temperature, 64)
-		gpu.MemoryUsed = bytesToMegabytes(memoryUsage)
-		gpu.MemoryTotal = bytesToMegabytes(totalMemory)
+		gpu.MemoryUsed = utils.BytesToMegabytes(memoryUsage)
+		gpu.MemoryTotal = utils.BytesToMegabytes(totalMemory)
 		gpu.Usage += usage
 		gpu.Power += power
 		gpu.Count++
@@ -320,16 +367,16 @@ func (gm *GPUManager) calculateGPUAverage(id string, gpu *system.GPUData, cacheK
 	gpuAvg := *gpu
 	deltaUsage, deltaPower, deltaPowerPkg := gm.calculateDeltas(gpu, lastSnapshot)
 
-	gpuAvg.Power = twoDecimals(deltaPower / float64(deltaCount))
+	gpuAvg.Power = utils.TwoDecimals(deltaPower / float64(deltaCount))
 
 	if gpu.Engines != nil {
 		// make fresh map for averaged engine metrics to avoid mutating
 		// the accumulator map stored in gm.GpuDataMap
 		gpuAvg.Engines = make(map[string]float64, len(gpu.Engines))
 		gpuAvg.Usage = gm.calculateIntelGPUUsage(&gpuAvg, gpu, lastSnapshot, deltaCount)
-		gpuAvg.PowerPkg = twoDecimals(deltaPowerPkg / float64(deltaCount))
+		gpuAvg.PowerPkg = utils.TwoDecimals(deltaPowerPkg / float64(deltaCount))
 	} else {
-		gpuAvg.Usage = twoDecimals(deltaUsage / float64(deltaCount))
+		gpuAvg.Usage = utils.TwoDecimals(deltaUsage / float64(deltaCount))
 	}
 
 	gm.lastAvgData[id] = gpuAvg
@@ -364,17 +411,17 @@ func (gm *GPUManager) calculateIntelGPUUsage(gpuAvg, gpu *system.GPUData, lastSn
 		} else {
 			deltaEngine = engine
 		}
-		gpuAvg.Engines[name] = twoDecimals(deltaEngine / float64(deltaCount))
+		gpuAvg.Engines[name] = utils.TwoDecimals(deltaEngine / float64(deltaCount))
 		maxEngineUsage = max(maxEngineUsage, deltaEngine/float64(deltaCount))
 	}
-	return twoDecimals(maxEngineUsage)
+	return utils.TwoDecimals(maxEngineUsage)
 }
 
 // updateInstantaneousValues updates values that should reflect current state, not averages
 func (gm *GPUManager) updateInstantaneousValues(gpuAvg *system.GPUData, gpu *system.GPUData) {
-	gpuAvg.Temperature = twoDecimals(gpu.Temperature)
-	gpuAvg.MemoryUsed = twoDecimals(gpu.MemoryUsed)
-	gpuAvg.MemoryTotal = twoDecimals(gpu.MemoryTotal)
+	gpuAvg.Temperature = utils.TwoDecimals(gpu.Temperature)
+	gpuAvg.MemoryUsed = utils.TwoDecimals(gpu.MemoryUsed)
+	gpuAvg.MemoryTotal = utils.TwoDecimals(gpu.MemoryTotal)
 }
 
 // storeSnapshot saves the current GPU state for this cache key
@@ -392,133 +439,325 @@ func (gm *GPUManager) storeSnapshot(id string, gpu *system.GPUData, cacheKey uin
 	gm.lastSnapshots[cacheKey][id] = snapshot
 }
 
-// detectGPUs checks for the presence of GPU management tools (nvidia-smi, rocm-smi, tegrastats)
-// in the system path. It sets the corresponding flags in the GPUManager struct if any of these
-// tools are found. If none of the tools are found, it returns an error indicating that no GPU
-// management tools are available.
-func (gm *GPUManager) detectGPUs() error {
+// discoverGpuCapabilities checks for available GPU tooling and sysfs support.
+// It only reports capability presence and does not apply policy decisions.
+func (gm *GPUManager) discoverGpuCapabilities() gpuCapabilities {
+	caps := gpuCapabilities{
+		hasAmdSysfs: gm.hasAmdSysfs(),
+	}
 	if _, err := exec.LookPath(nvidiaSmiCmd); err == nil {
-		gm.nvidiaSmi = true
+		caps.hasNvidiaSmi = true
 	}
 	if _, err := exec.LookPath(rocmSmiCmd); err == nil {
-		if val, _ := GetEnv("AMD_SYSFS"); val == "true" {
-			gm.amdgpu = true
-		} else {
-			gm.rocmSmi = true
-		}
-	} else if gm.hasAmdSysfs() {
-		gm.amdgpu = true
+		caps.hasRocmSmi = true
 	}
 	if _, err := exec.LookPath(tegraStatsCmd); err == nil {
-		gm.tegrastats = true
-		gm.nvidiaSmi = false
+		caps.hasTegrastats = true
 	}
 	if _, err := exec.LookPath(intelGpuStatsCmd); err == nil {
-		gm.intelGpuStats = true
+		caps.hasIntelGpuTop = true
 	}
-	if gm.nvidiaSmi || gm.rocmSmi || gm.amdgpu || gm.tegrastats || gm.intelGpuStats || gm.nvml {
-		return nil
+	if _, err := exec.LookPath(nvtopCmd); err == nil {
+		caps.hasNvtop = true
 	}
-	return fmt.Errorf("no GPU found - install nvidia-smi, rocm-smi, or intel_gpu_top")
+	if runtime.GOOS == "darwin" {
+		if _, err := utils.LookPathHomebrew(macmonCmd); err == nil {
+			caps.hasMacmon = true
+		}
+		if _, err := exec.LookPath(powermetricsCmd); err == nil {
+			caps.hasPowermetrics = true
+		}
+	}
+	return caps
 }
 
-// startCollector starts the appropriate GPU data collector based on the command
-func (gm *GPUManager) startCollector(command string) {
-	collector := gpuCollector{
-		name:    command,
-		bufSize: 10 * 1024,
-	}
-	switch command {
-	case intelGpuStatsCmd:
-		go func() {
-			failures := 0
-			for {
-				if err := gm.collectIntelStats(); err != nil {
-					failures++
-					if failures > maxFailureRetries {
-						break
-					}
-					slog.Warn("Error collecting Intel GPU data; see https://beszel.dev/guide/gpu", "err", err)
-					time.Sleep(retryWaitTime)
-					continue
+func hasAnyGpuCollector(caps gpuCapabilities) bool {
+	return caps.hasNvidiaSmi || caps.hasRocmSmi || caps.hasAmdSysfs || caps.hasTegrastats || caps.hasIntelGpuTop || caps.hasNvtop || caps.hasMacmon || caps.hasPowermetrics
+}
+
+func (gm *GPUManager) startIntelCollector() {
+	go func() {
+		failures := 0
+		for {
+			if err := gm.collectIntelStats(); err != nil {
+				failures++
+				if failures > maxFailureRetries {
+					break
 				}
+				slog.Warn("Error collecting Intel GPU data; see https://beszel.dev/guide/gpu", "err", err)
+				time.Sleep(retryWaitTime)
+				continue
 			}
-		}()
-	case nvidiaSmiCmd:
-		collector.cmdArgs = []string{
-			"-l", nvidiaSmiInterval,
+		}
+	}()
+}
+
+func (gm *GPUManager) startNvidiaSmiCollector(intervalSeconds string) {
+	collector := gpuCollector{
+		name:    nvidiaSmiCmd,
+		bufSize: 10 * 1024,
+		cmdArgs: []string{
+			"-l", intervalSeconds,
 			"--query-gpu=index,name,temperature.gpu,memory.used,memory.total,utilization.gpu,power.draw",
 			"--format=csv,noheader,nounits",
-		}
-		collector.parse = gm.parseNvidiaData
-		go collector.start()
-	case tegraStatsCmd:
-		collector.cmdArgs = []string{"--interval", tegraStatsInterval}
-		collector.parse = gm.getJetsonParser()
-		go collector.start()
-	case amdgpuCmd:
-		go func() {
-			if err := gm.collectAmdStats(); err != nil {
-				slog.Warn("Error collecting AMD GPU data via sysfs", "err", err)
-			}
-		}()
-	case rocmSmiCmd:
-		collector.cmdArgs = []string{"--showid", "--showtemp", "--showuse", "--showpower", "--showproductname", "--showmeminfo", "vram", "--json"}
-		collector.parse = gm.parseAmdData
-		go func() {
-			failures := 0
-			for {
-				if err := collector.collect(); err != nil {
-					failures++
-					if failures > maxFailureRetries {
-						break
-					}
-					slog.Warn("Error collecting AMD GPU data via rocm-smi", "err", err)
-				}
-				time.Sleep(rocmSmiInterval)
-			}
-		}()
+		},
+		parse: gm.parseNvidiaData,
 	}
+	go collector.start()
+}
+
+func (gm *GPUManager) startTegraStatsCollector(intervalMilliseconds string) {
+	collector := gpuCollector{
+		name:    tegraStatsCmd,
+		bufSize: 10 * 1024,
+		cmdArgs: []string{"--interval", intervalMilliseconds},
+		parse:   gm.getJetsonParser(),
+	}
+	go collector.start()
+}
+
+func (gm *GPUManager) startRocmSmiCollector(pollInterval time.Duration) {
+	collector := gpuCollector{
+		name:    rocmSmiCmd,
+		bufSize: 10 * 1024,
+		cmdArgs: []string{"--showid", "--showtemp", "--showuse", "--showpower", "--showproductname", "--showmeminfo", "vram", "--json"},
+		parse:   gm.parseAmdData,
+	}
+	go func() {
+		failures := 0
+		for {
+			if err := collector.collect(); err != nil {
+				failures++
+				if failures > maxFailureRetries {
+					break
+				}
+				slog.Warn("Error collecting AMD GPU data via rocm-smi", "err", err)
+			}
+			time.Sleep(pollInterval)
+		}
+	}()
+}
+
+func (gm *GPUManager) collectorDefinitions(caps gpuCapabilities) map[collectorSource]collectorDefinition {
+	return map[collectorSource]collectorDefinition{
+		collectorSourceNVML: {
+			group:     collectorGroupNvidia,
+			available: true,
+			start: func(_ func()) bool {
+				return gm.startNvmlCollector()
+			},
+		},
+		collectorSourceNvidiaSMI: {
+			group:     collectorGroupNvidia,
+			available: caps.hasNvidiaSmi,
+			start: func(_ func()) bool {
+				gm.startNvidiaSmiCollector("4") // seconds
+				return true
+			},
+		},
+		collectorSourceIntelGpuTop: {
+			group:     collectorGroupIntel,
+			available: caps.hasIntelGpuTop,
+			start: func(_ func()) bool {
+				gm.startIntelCollector()
+				return true
+			},
+		},
+		collectorSourceAmdSysfs: {
+			group:     collectorGroupAmd,
+			available: caps.hasAmdSysfs,
+			start: func(_ func()) bool {
+				return gm.startAmdSysfsCollector()
+			},
+		},
+		collectorSourceRocmSMI: {
+			group:              collectorGroupAmd,
+			available:          caps.hasRocmSmi,
+			deprecationWarning: "rocm-smi is deprecated and may be removed in a future release",
+			start: func(_ func()) bool {
+				gm.startRocmSmiCollector(4300 * time.Millisecond)
+				return true
+			},
+		},
+		collectorSourceNVTop: {
+			available: caps.hasNvtop,
+			start: func(onFailure func()) bool {
+				gm.startNvtopCollector("30", onFailure) // tens of milliseconds
+				return true
+			},
+		},
+		collectorSourceMacmon: {
+			group:     collectorGroupApple,
+			available: caps.hasMacmon,
+			start: func(_ func()) bool {
+				gm.startMacmonCollector()
+				return true
+			},
+		},
+		collectorSourcePowermetrics: {
+			group:     collectorGroupApple,
+			available: caps.hasPowermetrics,
+			start: func(_ func()) bool {
+				gm.startPowermetricsCollector()
+				return true
+			},
+		},
+	}
+}
+
+// parseCollectorPriority parses GPU_COLLECTOR and returns valid ordered entries.
+func parseCollectorPriority(value string) []collectorSource {
+	parts := strings.Split(value, ",")
+	priorities := make([]collectorSource, 0, len(parts))
+	for _, raw := range parts {
+		name := collectorSource(strings.TrimSpace(strings.ToLower(raw)))
+		if !isValidCollectorSource(name) {
+			if name != "" {
+				slog.Warn("Ignoring unknown GPU collector", "collector", name)
+			}
+			continue
+		}
+		priorities = append(priorities, name)
+	}
+	return priorities
+}
+
+// startNvmlCollector initializes NVML and starts its polling loop.
+func (gm *GPUManager) startNvmlCollector() bool {
+	collector := &nvmlCollector{gm: gm}
+	if err := collector.init(); err != nil {
+		slog.Warn("Failed to initialize NVML", "err", err)
+		return false
+	}
+	go collector.start()
+	return true
+}
+
+// startAmdSysfsCollector starts AMD GPU collection via sysfs.
+func (gm *GPUManager) startAmdSysfsCollector() bool {
+	go func() {
+		if err := gm.collectAmdStats(); err != nil {
+			slog.Warn("Error collecting AMD GPU data via sysfs", "err", err)
+		}
+	}()
+	return true
+}
+
+// startCollectorsByPriority starts collectors in order with one source per vendor group.
+func (gm *GPUManager) startCollectorsByPriority(priorities []collectorSource, caps gpuCapabilities) int {
+	definitions := gm.collectorDefinitions(caps)
+	selectedGroups := make(map[string]bool, 3)
+	started := 0
+	for i, source := range priorities {
+		definition, ok := definitions[source]
+		if !ok || !definition.available {
+			continue
+		}
+		// nvtop is not a vendor-specific collector, so should only be used if no other collectors are selected or it is first in GPU_COLLECTOR.
+		if source == collectorSourceNVTop {
+			if len(selectedGroups) > 0 {
+				slog.Warn("Skipping nvtop because other collectors are selected")
+				continue
+			}
+			// if nvtop fails, fall back to remaining collectors.
+			remaining := append([]collectorSource(nil), priorities[i+1:]...)
+			if definition.start(func() {
+				gm.startCollectorsByPriority(remaining, caps)
+			}) {
+				started++
+				return started
+			}
+		}
+		group := definition.group
+		if group == "" || selectedGroups[group] {
+			continue
+		}
+		if definition.deprecationWarning != "" {
+			slog.Warn(definition.deprecationWarning)
+		}
+		if definition.start(nil) {
+			selectedGroups[group] = true
+			started++
+		}
+	}
+	return started
+}
+
+// resolveLegacyCollectorPriority builds the default collector order when GPU_COLLECTOR is unset.
+func (gm *GPUManager) resolveLegacyCollectorPriority(caps gpuCapabilities) []collectorSource {
+	priorities := make([]collectorSource, 0, 4)
+
+	if caps.hasNvidiaSmi && !caps.hasTegrastats {
+		if nvml, _ := utils.GetEnv("NVML"); nvml == "true" {
+			priorities = append(priorities, collectorSourceNVML, collectorSourceNvidiaSMI)
+		} else {
+			priorities = append(priorities, collectorSourceNvidiaSMI)
+		}
+	}
+
+	if caps.hasRocmSmi {
+		if val, _ := utils.GetEnv("AMD_SYSFS"); val == "true" {
+			priorities = append(priorities, collectorSourceAmdSysfs)
+		} else {
+			priorities = append(priorities, collectorSourceRocmSMI)
+		}
+	} else if caps.hasAmdSysfs {
+		priorities = append(priorities, collectorSourceAmdSysfs)
+	}
+
+	if caps.hasIntelGpuTop {
+		priorities = append(priorities, collectorSourceIntelGpuTop)
+	}
+
+	// Apple collectors are currently opt-in only for testing.
+	// Enable them with GPU_COLLECTOR=macmon or GPU_COLLECTOR=powermetrics.
+	// TODO: uncomment below when Apple collectors are confirmed to be working.
+	//
+	// Prefer macmon on macOS (no sudo). Fall back to powermetrics if present.
+	// if caps.hasMacmon {
+	// 	priorities = append(priorities, collectorSourceMacmon)
+	// } else if caps.hasPowermetrics {
+	// 	priorities = append(priorities, collectorSourcePowermetrics)
+	// }
+
+	// Keep nvtop as a last resort only when no vendor collector exists.
+	if len(priorities) == 0 && caps.hasNvtop {
+		priorities = append(priorities, collectorSourceNVTop)
+	}
+	return priorities
 }
 
 // NewGPUManager creates and initializes a new GPUManager
 func NewGPUManager() (*GPUManager, error) {
-	if skipGPU, _ := GetEnv("SKIP_GPU"); skipGPU == "true" {
+	if skipGPU, _ := utils.GetEnv("SKIP_GPU"); skipGPU == "true" {
 		return nil, nil
 	}
 	var gm GPUManager
-	if err := gm.detectGPUs(); err != nil {
-		return nil, err
-	}
+	caps := gm.discoverGpuCapabilities()
 	gm.GpuDataMap = make(map[string]*system.GPUData)
 
-	if gm.nvidiaSmi {
-		if nvml, _ := GetEnv("NVML"); nvml == "true" {
-			gm.nvml = true
-			gm.nvidiaSmi = false
-			collector := &nvmlCollector{gm: &gm}
-			if err := collector.init(); err == nil {
-				go collector.start()
-			} else {
-				slog.Warn("Failed to initialize NVML, falling back to nvidia-smi", "err", err)
-				gm.nvidiaSmi = true
-				gm.startCollector(nvidiaSmiCmd)
-			}
-		} else {
-			gm.startCollector(nvidiaSmiCmd)
+	// Jetson devices should always use tegrastats (ignore GPU_COLLECTOR).
+	if caps.hasTegrastats {
+		gm.startTegraStatsCollector("3700")
+		return &gm, nil
+	}
+
+	// Respect explicit collector selection before capability auto-detection.
+	if collectorConfig, ok := utils.GetEnv("GPU_COLLECTOR"); ok && strings.TrimSpace(collectorConfig) != "" {
+		priorities := parseCollectorPriority(collectorConfig)
+		if gm.startCollectorsByPriority(priorities, caps) == 0 {
+			return nil, fmt.Errorf("no configured GPU collectors are available")
 		}
+		return &gm, nil
 	}
-	if gm.rocmSmi {
-		gm.startCollector(rocmSmiCmd)
+
+	if !hasAnyGpuCollector(caps) {
+		return nil, fmt.Errorf(noGPUFoundMsg)
 	}
-	if gm.amdgpu {
-		gm.startCollector(amdgpuCmd)
-	}
-	if gm.tegrastats {
-		gm.startCollector(tegraStatsCmd)
-	}
-	if gm.intelGpuStats {
-		gm.startCollector(intelGpuStatsCmd)
+
+	// auto-detect and start collectors when GPU_COLLECTOR is unset.
+	if gm.startCollectorsByPriority(gm.resolveLegacyCollectorPriority(caps), caps) == 0 {
+		return nil, fmt.Errorf(noGPUFoundMsg)
 	}
 
 	return &gm, nil

agent/gpu_amd_linux.go

@@ -3,6 +3,7 @@
 package agent
 
 import (
+	"bufio"
 	"fmt"
 	"log/slog"
 	"os"
@@ -12,9 +13,19 @@ import (
 	"sync"
 	"time"
 
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"
 )
 
+var amdgpuNameCache = struct {
+	sync.RWMutex
+	hits   map[string]string
+	misses map[string]struct{}
+}{
+	hits:   make(map[string]string),
+	misses: make(map[string]struct{}),
+}
+
 // hasAmdSysfs returns true if any AMD GPU sysfs nodes are found
 func (gm *GPUManager) hasAmdSysfs() bool {
 	cards, err := filepath.Glob("/sys/class/drm/card*/device/vendor")
@@ -22,8 +33,8 @@ func (gm *GPUManager) hasAmdSysfs() bool {
 		return false
 	}
 	for _, vendorPath := range cards {
-		vendor, err := os.ReadFile(vendorPath)
-		if err == nil && strings.TrimSpace(string(vendor)) == "0x1002" {
+		vendor, err := utils.ReadStringFileLimited(vendorPath, 64)
+		if err == nil && vendor == "0x1002" {
 			return true
 		}
 	}
@@ -32,6 +43,7 @@ func (gm *GPUManager) hasAmdSysfs() bool {
 
 // collectAmdStats collects AMD GPU metrics directly from sysfs to avoid the overhead of rocm-smi
 func (gm *GPUManager) collectAmdStats() error {
+	sysfsPollInterval := 3000 * time.Millisecond
 	cards, err := filepath.Glob("/sys/class/drm/card*")
 	if err != nil {
 		return err
@@ -70,17 +82,17 @@ func (gm *GPUManager) collectAmdStats() error {
 			continue
 		}
 		failures = 0
-		time.Sleep(rocmSmiInterval)
+		time.Sleep(sysfsPollInterval)
 	}
 }
 
+// isAmdGpu checks whether a DRM card path belongs to AMD vendor ID 0x1002.
 func isAmdGpu(cardPath string) bool {
-	vendorPath := filepath.Join(cardPath, "device/vendor")
-	vendor, err := os.ReadFile(vendorPath)
+	vendor, err := utils.ReadStringFileLimited(filepath.Join(cardPath, "device/vendor"), 64)
 	if err != nil {
 		return false
 	}
-	return strings.TrimSpace(string(vendor)) == "0x1002"
+	return vendor == "0x1002"
 }
 
 // updateAmdGpuData reads GPU metrics from sysfs and updates the GPU data map.
@@ -93,6 +105,13 @@ func (gm *GPUManager) updateAmdGpuData(cardPath string) bool {
 	usage, usageErr := readSysfsFloat(filepath.Join(devicePath, "gpu_busy_percent"))
 	memUsed, memUsedErr := readSysfsFloat(filepath.Join(devicePath, "mem_info_vram_used"))
 	memTotal, _ := readSysfsFloat(filepath.Join(devicePath, "mem_info_vram_total"))
+	// if gtt is present, add it to the memory used and total (https://github.com/henrygd/beszel/issues/1569#issuecomment-3837640484)
+	if gttUsed, err := readSysfsFloat(filepath.Join(devicePath, "mem_info_gtt_used")); err == nil && gttUsed > 0 {
+		if gttTotal, err := readSysfsFloat(filepath.Join(devicePath, "mem_info_gtt_total")); err == nil {
+			memUsed += gttUsed
+			memTotal += gttTotal
+		}
+	}
 
 	var temp, power float64
 	hwmons, _ := filepath.Glob(filepath.Join(devicePath, "hwmon/hwmon*"))
@@ -125,20 +144,129 @@ func (gm *GPUManager) updateAmdGpuData(cardPath string) bool {
 	if usageErr == nil {
 		gpu.Usage += usage
 	}
-	gpu.MemoryUsed = bytesToMegabytes(memUsed)
-	gpu.MemoryTotal = bytesToMegabytes(memTotal)
+	gpu.MemoryUsed = utils.BytesToMegabytes(memUsed)
+	gpu.MemoryTotal = utils.BytesToMegabytes(memTotal)
 	gpu.Temperature = temp
 	gpu.Power += power
 	gpu.Count++
 	return true
 }
 
+// readSysfsFloat reads and parses a numeric value from a sysfs file.
 func readSysfsFloat(path string) (float64, error) {
-	val, err := os.ReadFile(path)
+	val, err := utils.ReadStringFileLimited(path, 64)
 	if err != nil {
+		slog.Debug("Failed to read sysfs value", "path", path, "error", err)
 		return 0, err
 	}
-	return strconv.ParseFloat(strings.TrimSpace(string(val)), 64)
+	return strconv.ParseFloat(val, 64)
+}
+
+// normalizeHexID normalizes hex IDs by trimming spaces, lowercasing, and dropping 0x.
+func normalizeHexID(id string) string {
+	return strings.TrimPrefix(strings.ToLower(strings.TrimSpace(id)), "0x")
+}
+
+// cacheKeyForAmdgpu builds the cache key for a device and optional revision.
+func cacheKeyForAmdgpu(deviceID, revisionID string) string {
+	if revisionID != "" {
+		return deviceID + ":" + revisionID
+	}
+	return deviceID
+}
+
+// lookupAmdgpuNameInFile resolves an AMDGPU name from amdgpu.ids by device/revision.
+func lookupAmdgpuNameInFile(deviceID, revisionID, filePath string) (name string, exact bool, found bool) {
+	file, err := os.Open(filePath)
+	if err != nil {
+		return "", false, false
+	}
+	defer file.Close()
+
+	var byDevice string
+	scanner := bufio.NewScanner(file)
+	for scanner.Scan() {
+		line := strings.TrimSpace(scanner.Text())
+		if line == "" || strings.HasPrefix(line, "#") {
+			continue
+		}
+		parts := strings.SplitN(line, ",", 3)
+		if len(parts) != 3 {
+			continue
+		}
+
+		dev := normalizeHexID(parts[0])
+		rev := normalizeHexID(parts[1])
+		productName := strings.TrimSpace(parts[2])
+		if dev == "" || productName == "" || dev != deviceID {
+			continue
+		}
+		if byDevice == "" {
+			byDevice = productName
+		}
+		if revisionID != "" && rev == revisionID {
+			return productName, true, true
+		}
+	}
+	if byDevice != "" {
+		return byDevice, false, true
+	}
+	return "", false, false
+}
+
+// getCachedAmdgpuName returns cached hit/miss status for the given device/revision.
+func getCachedAmdgpuName(deviceID, revisionID string) (name string, found bool, done bool) {
+	// Build the list of cache keys to check. We always look up the exact device+revision key.
+	// When revisionID is set, we also look up deviceID alone, since the cache may store a
+	// device-only fallback when we couldn't resolve the exact revision.
+	keys := []string{cacheKeyForAmdgpu(deviceID, revisionID)}
+	if revisionID != "" {
+		keys = append(keys, deviceID)
+	}
+
+	knownMisses := 0
+	amdgpuNameCache.RLock()
+	defer amdgpuNameCache.RUnlock()
+	for _, key := range keys {
+		if name, ok := amdgpuNameCache.hits[key]; ok {
+			return name, true, true
+		}
+		if _, ok := amdgpuNameCache.misses[key]; ok {
+			knownMisses++
+		}
+	}
+	// done=true means "don't bother doing slow lookup": we either found a name (above) or
+	// every key we checked was already a known miss, so we've tried before and failed.
+	return "", false, knownMisses == len(keys)
+}
+
+// normalizeAmdgpuName trims standard suffixes from AMDGPU product names.
+func normalizeAmdgpuName(name string) string {
+	for _, suffix := range []string{" Graphics", " Series"} {
+		name = strings.TrimSuffix(name, suffix)
+	}
+	return name
+}
+
+// cacheAmdgpuName stores a resolved AMDGPU name in the lookup cache.
+func cacheAmdgpuName(deviceID, revisionID, name string, exact bool) {
+	name = normalizeAmdgpuName(name)
+	amdgpuNameCache.Lock()
+	defer amdgpuNameCache.Unlock()
+	if exact && revisionID != "" {
+		amdgpuNameCache.hits[cacheKeyForAmdgpu(deviceID, revisionID)] = name
+	}
+	amdgpuNameCache.hits[deviceID] = name
+}
+
+// cacheMissingAmdgpuName records unresolved device/revision lookups.
+func cacheMissingAmdgpuName(deviceID, revisionID string) {
+	amdgpuNameCache.Lock()
+	defer amdgpuNameCache.Unlock()
+	amdgpuNameCache.misses[deviceID] = struct{}{}
+	if revisionID != "" {
+		amdgpuNameCache.misses[cacheKeyForAmdgpu(deviceID, revisionID)] = struct{}{}
+	}
 }
 
 // getAmdGpuName attempts to get a descriptive GPU name.
@@ -146,39 +274,30 @@ func readSysfsFloat(path string) (float64, error) {
 // Falls back to showing the raw device ID if not found in the lookup table.
 func getAmdGpuName(devicePath string) string {
 	// Try product_name first (works for some enterprise GPUs)
-	if prod, err := os.ReadFile(filepath.Join(devicePath, "product_name")); err == nil {
-		return strings.TrimSpace(string(prod))
+	if prod, err := utils.ReadStringFileLimited(filepath.Join(devicePath, "product_name"), 128); err == nil {
+		return prod
 	}
 
 	// Read PCI device ID and look it up
-	if deviceID, err := os.ReadFile(filepath.Join(devicePath, "device")); err == nil {
-		id := strings.TrimPrefix(strings.ToLower(strings.TrimSpace(string(deviceID))), "0x")
-		if name, ok := getRadeonNames()[id]; ok {
-			return fmt.Sprintf("Radeon %s", name)
+	if deviceID, err := utils.ReadStringFileLimited(filepath.Join(devicePath, "device"), 64); err == nil {
+		id := normalizeHexID(deviceID)
+		revision := ""
+		if rev, revErr := utils.ReadStringFileLimited(filepath.Join(devicePath, "revision"), 64); revErr == nil {
+			revision = normalizeHexID(rev)
 		}
+
+		if name, found, done := getCachedAmdgpuName(id, revision); found {
+			return name
+		} else if !done {
+			if name, exact, ok := lookupAmdgpuNameInFile(id, revision, "/usr/share/libdrm/amdgpu.ids"); ok {
+				cacheAmdgpuName(id, revision, name, exact)
+				return normalizeAmdgpuName(name)
+			}
+			cacheMissingAmdgpuName(id, revision)
+		}
+
 		return fmt.Sprintf("AMD GPU (%s)", id)
 	}
 
 	return "AMD GPU"
 }
-
-// getRadeonNames returns the AMD GPU name lookup table
-// Device IDs from https://pci-ids.ucw.cz/read/PC/1002
-var getRadeonNames = sync.OnceValue(func() map[string]string {
-	return map[string]string{
-		"7550": "RX 9070",
-		"7590": "RX 9060 XT",
-		"7551": "AI PRO R9700",
-
-		"744c": "RX 7900",
-
-		"1681": "680M",
-
-		"7448": "PRO W7900",
-		"745e": "PRO W7800",
-		"7470": "PRO W7700",
-		"73e3": "PRO W6600",
-		"7422": "PRO W6400",
-		"7341": "PRO W5500",
-	}
-})

agent/gpu_amd_linux_test.go

@@ -0,0 +1,265 @@
+//go:build linux
+
+package agent
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/henrygd/beszel/agent/utils"
+	"github.com/henrygd/beszel/internal/entities/system"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestNormalizeHexID(t *testing.T) {
+	tests := []struct {
+		in   string
+		want string
+	}{
+		{"0x1002", "1002"},
+		{"C2", "c2"},
+		{"  15BF  ", "15bf"},
+		{"0x15bf", "15bf"},
+		{"", ""},
+	}
+	for _, tt := range tests {
+		subName := tt.in
+		if subName == "" {
+			subName = "empty_string"
+		}
+		t.Run(subName, func(t *testing.T) {
+			got := normalizeHexID(tt.in)
+			assert.Equal(t, tt.want, got)
+		})
+	}
+}
+
+func TestCacheKeyForAmdgpu(t *testing.T) {
+	tests := []struct {
+		deviceID   string
+		revisionID string
+		want       string
+	}{
+		{"1114", "c2", "1114:c2"},
+		{"15bf", "", "15bf"},
+		{"1506", "c1", "1506:c1"},
+	}
+	for _, tt := range tests {
+		got := cacheKeyForAmdgpu(tt.deviceID, tt.revisionID)
+		assert.Equal(t, tt.want, got)
+	}
+}
+
+func TestReadSysfsFloat(t *testing.T) {
+	dir := t.TempDir()
+
+	validPath := filepath.Join(dir, "val")
+	require.NoError(t, os.WriteFile(validPath, []byte("  42.5  \n"), 0o644))
+	got, err := readSysfsFloat(validPath)
+	require.NoError(t, err)
+	assert.Equal(t, 42.5, got)
+
+	// Integer and scientific
+	sciPath := filepath.Join(dir, "sci")
+	require.NoError(t, os.WriteFile(sciPath, []byte("1e2"), 0o644))
+	got, err = readSysfsFloat(sciPath)
+	require.NoError(t, err)
+	assert.Equal(t, 100.0, got)
+
+	// Missing file
+	_, err = readSysfsFloat(filepath.Join(dir, "missing"))
+	require.Error(t, err)
+
+	// Invalid content
+	badPath := filepath.Join(dir, "bad")
+	require.NoError(t, os.WriteFile(badPath, []byte("not a number"), 0o644))
+	_, err = readSysfsFloat(badPath)
+	require.Error(t, err)
+}
+
+func TestIsAmdGpu(t *testing.T) {
+	dir := t.TempDir()
+	deviceDir := filepath.Join(dir, "device")
+	require.NoError(t, os.MkdirAll(deviceDir, 0o755))
+
+	// AMD vendor 0x1002 -> true
+	require.NoError(t, os.WriteFile(filepath.Join(deviceDir, "vendor"), []byte("0x1002\n"), 0o644))
+	assert.True(t, isAmdGpu(dir), "vendor 0x1002 should be AMD")
+
+	// Non-AMD vendor -> false
+	require.NoError(t, os.WriteFile(filepath.Join(deviceDir, "vendor"), []byte("0x10de\n"), 0o644))
+	assert.False(t, isAmdGpu(dir), "vendor 0x10de should not be AMD")
+
+	// Missing vendor file -> false
+	require.NoError(t, os.Remove(filepath.Join(deviceDir, "vendor")))
+	assert.False(t, isAmdGpu(dir), "missing vendor file should be false")
+}
+
+func TestAmdgpuNameCacheRoundTrip(t *testing.T) {
+	// Cache a name and retrieve it (unique key to avoid affecting other tests)
+	deviceID, revisionID := "cachedev99", "00"
+	cacheAmdgpuName(deviceID, revisionID, "AMD Test GPU 99 Graphics", true)
+
+	name, found, done := getCachedAmdgpuName(deviceID, revisionID)
+	assert.True(t, found)
+	assert.True(t, done)
+	assert.Equal(t, "AMD Test GPU 99", name)
+
+	// Device-only key also stored
+	name2, found2, _ := getCachedAmdgpuName(deviceID, "")
+	assert.True(t, found2)
+	assert.Equal(t, "AMD Test GPU 99", name2)
+
+	// Cache a miss
+	cacheMissingAmdgpuName("missedev99", "ab")
+	_, found3, done3 := getCachedAmdgpuName("missedev99", "ab")
+	assert.False(t, found3)
+	assert.True(t, done3, "done should be true so caller skips file lookup")
+}
+
+func TestUpdateAmdGpuDataWithFakeSysfs(t *testing.T) {
+	tests := []struct {
+		name            string
+		writeGTT        bool
+		wantMemoryUsed  float64
+		wantMemoryTotal float64
+	}{
+		{
+			name:            "sums vram and gtt when gtt is present",
+			writeGTT:        true,
+			wantMemoryUsed:  utils.BytesToMegabytes(1073741824 + 536870912),
+			wantMemoryTotal: utils.BytesToMegabytes(2147483648 + 4294967296),
+		},
+		{
+			name:            "falls back to vram when gtt is missing",
+			writeGTT:        false,
+			wantMemoryUsed:  utils.BytesToMegabytes(1073741824),
+			wantMemoryTotal: utils.BytesToMegabytes(2147483648),
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			dir := t.TempDir()
+			cardPath := filepath.Join(dir, "card0")
+			devicePath := filepath.Join(cardPath, "device")
+			hwmonPath := filepath.Join(devicePath, "hwmon", "hwmon0")
+			require.NoError(t, os.MkdirAll(hwmonPath, 0o755))
+
+			write := func(name, content string) {
+				require.NoError(t, os.WriteFile(filepath.Join(devicePath, name), []byte(content), 0o644))
+			}
+			write("vendor", "0x1002")
+			write("device", "0x1506")
+			write("revision", "0xc1")
+			write("gpu_busy_percent", "25")
+			write("mem_info_vram_used", "1073741824")
+			write("mem_info_vram_total", "2147483648")
+			if tt.writeGTT {
+				write("mem_info_gtt_used", "536870912")
+				write("mem_info_gtt_total", "4294967296")
+			}
+			require.NoError(t, os.WriteFile(filepath.Join(hwmonPath, "temp1_input"), []byte("45000"), 0o644))
+			require.NoError(t, os.WriteFile(filepath.Join(hwmonPath, "power1_input"), []byte("20000000"), 0o644))
+
+			// Pre-cache name so getAmdGpuName returns a known value (it uses system amdgpu.ids path)
+			cacheAmdgpuName("1506", "c1", "AMD Radeon 610M Graphics", true)
+
+			gm := &GPUManager{GpuDataMap: make(map[string]*system.GPUData)}
+			ok := gm.updateAmdGpuData(cardPath)
+			require.True(t, ok)
+
+			gpu, ok := gm.GpuDataMap["card0"]
+			require.True(t, ok)
+			assert.Equal(t, "AMD Radeon 610M", gpu.Name)
+			assert.Equal(t, 25.0, gpu.Usage)
+			assert.Equal(t, tt.wantMemoryUsed, gpu.MemoryUsed)
+			assert.Equal(t, tt.wantMemoryTotal, gpu.MemoryTotal)
+			assert.Equal(t, 45.0, gpu.Temperature)
+			assert.Equal(t, 20.0, gpu.Power)
+			assert.Equal(t, 1.0, gpu.Count)
+		})
+	}
+}
+
+func TestLookupAmdgpuNameInFile(t *testing.T) {
+	idsPath := filepath.Join("test-data", "amdgpu.ids")
+
+	tests := []struct {
+		name       string
+		deviceID   string
+		revisionID string
+		wantName   string
+		wantExact  bool
+		wantFound  bool
+	}{
+		{
+			name:       "exact device and revision match",
+			deviceID:   "1114",
+			revisionID: "c2",
+			wantName:   "AMD Radeon 860M Graphics",
+			wantExact:  true,
+			wantFound:  true,
+		},
+		{
+			name:       "exact match 15BF revision 01 returns 760M",
+			deviceID:   "15bf",
+			revisionID: "01",
+			wantName:   "AMD Radeon 760M Graphics",
+			wantExact:  true,
+			wantFound:  true,
+		},
+		{
+			name:       "exact match 15BF revision 00 returns 780M",
+			deviceID:   "15bf",
+			revisionID: "00",
+			wantName:   "AMD Radeon 780M Graphics",
+			wantExact:  true,
+			wantFound:  true,
+		},
+		{
+			name:       "device-only match returns first entry for device",
+			deviceID:   "1506",
+			revisionID: "",
+			wantName:   "AMD Radeon 610M",
+			wantExact:  false,
+			wantFound:  true,
+		},
+		{
+			name:       "unknown device not found",
+			deviceID:   "dead",
+			revisionID: "00",
+			wantName:   "",
+			wantExact:  false,
+			wantFound:  false,
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			gotName, gotExact, gotFound := lookupAmdgpuNameInFile(tt.deviceID, tt.revisionID, idsPath)
+			assert.Equal(t, tt.wantName, gotName, "name")
+			assert.Equal(t, tt.wantExact, gotExact, "exact")
+			assert.Equal(t, tt.wantFound, gotFound, "found")
+		})
+	}
+}
+
+func TestGetAmdGpuNameFromIdsFile(t *testing.T) {
+	// Test that getAmdGpuName resolves a name when we can't inject the ids path.
+	// We only verify behavior when product_name is missing and device/revision
+	// would be read from sysfs; the actual lookup uses /usr/share/libdrm/amdgpu.ids.
+	// So this test focuses on normalizeAmdgpuName and that lookupAmdgpuNameInFile
+	// returns the expected name for our test-data file.
+	idsPath := filepath.Join("test-data", "amdgpu.ids")
+	name, exact, found := lookupAmdgpuNameInFile("1435", "ae", idsPath)
+	require.True(t, found)
+	require.True(t, exact)
+	assert.Equal(t, "AMD Custom GPU 0932", name)
+	assert.Equal(t, "AMD Custom GPU 0932", normalizeAmdgpuName(name))
+
+	// " Graphics" suffix is trimmed by normalizeAmdgpuName
+	name2 := "AMD Radeon 860M Graphics"
+	assert.Equal(t, "AMD Radeon 860M", normalizeAmdgpuName(name2))
+}

agent/gpu_darwin.go

@@ -0,0 +1,257 @@
+//go:build darwin
+
+package agent
+
+import (
+	"bufio"
+	"bytes"
+	"encoding/json"
+	"io"
+	"log/slog"
+	"os/exec"
+	"strconv"
+	"strings"
+	"time"
+
+	"github.com/henrygd/beszel/agent/utils"
+	"github.com/henrygd/beszel/internal/entities/system"
+)
+
+const (
+	// powermetricsSampleIntervalMs is the sampling interval passed to powermetrics (-i).
+	powermetricsSampleIntervalMs = 500
+	// powermetricsPollInterval is how often we run powermetrics to collect a new sample.
+	powermetricsPollInterval = 2 * time.Second
+	// macmonIntervalMs is the sampling interval passed to macmon pipe (-i), in milliseconds.
+	macmonIntervalMs = 2500
+)
+
+const appleGPUID = "0"
+
+// startPowermetricsCollector runs powermetrics --samplers gpu_power in a loop and updates
+// GPU usage and power. Requires root (sudo) on macOS. A single logical GPU is reported as id "0".
+func (gm *GPUManager) startPowermetricsCollector() {
+	// Ensure single GPU entry for Apple GPU
+	if _, ok := gm.GpuDataMap[appleGPUID]; !ok {
+		gm.GpuDataMap[appleGPUID] = &system.GPUData{Name: "Apple GPU"}
+	}
+
+	go func() {
+		failures := 0
+		for {
+			if err := gm.collectPowermetrics(); err != nil {
+				failures++
+				if failures > maxFailureRetries {
+					slog.Warn("powermetrics GPU collector failed repeatedly, stopping", "err", err)
+					break
+				}
+				slog.Warn("Error collecting macOS GPU data via powermetrics (may require sudo)", "err", err)
+				time.Sleep(retryWaitTime)
+				continue
+			}
+			failures = 0
+			time.Sleep(powermetricsPollInterval)
+		}
+	}()
+}
+
+// collectPowermetrics runs powermetrics once and parses GPU usage and power from its output.
+func (gm *GPUManager) collectPowermetrics() error {
+	interval := strconv.Itoa(powermetricsSampleIntervalMs)
+	cmd := exec.Command(powermetricsCmd, "--samplers", "gpu_power", "-i", interval, "-n", "1")
+	cmd.Stderr = nil
+	out, err := cmd.Output()
+	if err != nil {
+		return err
+	}
+	if !gm.parsePowermetricsData(out) {
+		return errNoValidData
+	}
+	return nil
+}
+
+// parsePowermetricsData parses powermetrics gpu_power output and updates GpuDataMap["0"].
+// Example output:
+//
+//	**** GPU usage ****
+//	GPU HW active frequency: 444 MHz
+//	GPU HW active residency:   0.97% (444 MHz: .97% ...
+//	GPU idle residency:  99.03%
+//	GPU Power: 4 mW
+func (gm *GPUManager) parsePowermetricsData(output []byte) bool {
+	var idleResidency, powerMW float64
+	var gotIdle, gotPower bool
+
+	scanner := bufio.NewScanner(bytes.NewReader(output))
+	for scanner.Scan() {
+		line := strings.TrimSpace(scanner.Text())
+		if strings.HasPrefix(line, "GPU idle residency:") {
+			// "GPU idle residency:  99.03%"
+			fields := strings.Fields(strings.TrimPrefix(line, "GPU idle residency:"))
+			if len(fields) >= 1 {
+				pct := strings.TrimSuffix(fields[0], "%")
+				if v, err := strconv.ParseFloat(pct, 64); err == nil {
+					idleResidency = v
+					gotIdle = true
+				}
+			}
+		} else if strings.HasPrefix(line, "GPU Power:") {
+			// "GPU Power: 4 mW"
+			fields := strings.Fields(strings.TrimPrefix(line, "GPU Power:"))
+			if len(fields) >= 1 {
+				if v, err := strconv.ParseFloat(fields[0], 64); err == nil {
+					powerMW = v
+					gotPower = true
+				}
+			}
+		}
+	}
+	if err := scanner.Err(); err != nil {
+		return false
+	}
+	if !gotIdle && !gotPower {
+		return false
+	}
+
+	gm.Lock()
+	defer gm.Unlock()
+
+	if _, ok := gm.GpuDataMap[appleGPUID]; !ok {
+		gm.GpuDataMap[appleGPUID] = &system.GPUData{Name: "Apple GPU"}
+	}
+	gpu := gm.GpuDataMap[appleGPUID]
+
+	if gotIdle {
+		// Usage = 100 - idle residency (e.g. 100 - 99.03 = 0.97%)
+		gpu.Usage += 100 - idleResidency
+	}
+	if gotPower {
+		// mW -> W
+		gpu.Power += powerMW / milliwattsInAWatt
+	}
+	gpu.Count++
+	return true
+}
+
+// startMacmonCollector runs `macmon pipe` in a loop and parses one JSON object per line.
+// This collector does not require sudo. A single logical GPU is reported as id "0".
+func (gm *GPUManager) startMacmonCollector() {
+	if _, ok := gm.GpuDataMap[appleGPUID]; !ok {
+		gm.GpuDataMap[appleGPUID] = &system.GPUData{Name: "Apple GPU"}
+	}
+
+	go func() {
+		failures := 0
+		for {
+			if err := gm.collectMacmonPipe(); err != nil {
+				failures++
+				if failures > maxFailureRetries {
+					slog.Warn("macmon GPU collector failed repeatedly, stopping", "err", err)
+					break
+				}
+				slog.Warn("Error collecting macOS GPU data via macmon", "err", err)
+				time.Sleep(retryWaitTime)
+				continue
+			}
+			failures = 0
+			// `macmon pipe` is long-running; if it returns, wait a bit before restarting.
+			time.Sleep(retryWaitTime)
+		}
+	}()
+}
+
+type macmonTemp struct {
+	GPUTempAvg float64 `json:"gpu_temp_avg"`
+}
+
+type macmonSample struct {
+	GPUPower    float64    `json:"gpu_power"`     // watts (macmon reports fractional values)
+	GPURAMPower float64    `json:"gpu_ram_power"` // watts
+	GPUUsage    []float64  `json:"gpu_usage"`     // [freq_mhz, usage] where usage is typically 0..1
+	Temp        macmonTemp `json:"temp"`
+}
+
+func (gm *GPUManager) collectMacmonPipe() (err error) {
+	macmonPath, err := utils.LookPathHomebrew(macmonCmd)
+	if err != nil {
+		return err
+	}
+	cmd := exec.Command(macmonPath, "pipe", "-i", strconv.Itoa(macmonIntervalMs))
+	// Avoid blocking if macmon writes to stderr.
+	cmd.Stderr = io.Discard
+	stdout, err := cmd.StdoutPipe()
+	if err != nil {
+		return err
+	}
+	if err := cmd.Start(); err != nil {
+		return err
+	}
+
+	// Ensure we always reap the child to avoid zombies on any return path and
+	// propagate a non-zero exit code if no other error was set.
+	defer func() {
+		_ = stdout.Close()
+		if cmd.ProcessState == nil || !cmd.ProcessState.Exited() {
+			_ = cmd.Process.Kill()
+		}
+		if waitErr := cmd.Wait(); err == nil && waitErr != nil {
+			err = waitErr
+		}
+	}()
+
+	scanner := bufio.NewScanner(stdout)
+	var hadSample bool
+	for scanner.Scan() {
+		line := bytes.TrimSpace(scanner.Bytes())
+		if len(line) == 0 {
+			continue
+		}
+		if gm.parseMacmonLine(line) {
+			hadSample = true
+		}
+	}
+	if scanErr := scanner.Err(); scanErr != nil {
+		return scanErr
+	}
+	if !hadSample {
+		return errNoValidData
+	}
+	return nil
+}
+
+// parseMacmonLine parses a single macmon JSON line and updates Apple GPU metrics.
+func (gm *GPUManager) parseMacmonLine(line []byte) bool {
+	var sample macmonSample
+	if err := json.Unmarshal(line, &sample); err != nil {
+		return false
+	}
+
+	usage := 0.0
+	if len(sample.GPUUsage) >= 2 {
+		usage = sample.GPUUsage[1]
+		// Heuristic: macmon typically reports 0..1; convert to percentage.
+		if usage <= 1.0 {
+			usage *= 100
+		}
+	}
+
+	// Consider the line valid if it contains at least one GPU metric.
+	if usage == 0 && sample.GPUPower == 0 && sample.Temp.GPUTempAvg == 0 {
+		return false
+	}
+
+	gm.Lock()
+	defer gm.Unlock()
+
+	gpu, ok := gm.GpuDataMap[appleGPUID]
+	if !ok {
+		gpu = &system.GPUData{Name: "Apple GPU"}
+		gm.GpuDataMap[appleGPUID] = gpu
+	}
+	gpu.Temperature = sample.Temp.GPUTempAvg
+	gpu.Usage += usage
+	// macmon reports power in watts; include VRAM power if present.
+	gpu.Power += sample.GPUPower + sample.GPURAMPower
+	gpu.Count++
+	return true
+}

agent/gpu_darwin_test.go

@@ -0,0 +1,81 @@
+//go:build darwin
+
+package agent
+
+import (
+	"testing"
+
+	"github.com/henrygd/beszel/internal/entities/system"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestParsePowermetricsData(t *testing.T) {
+	input := `
+Machine model: Mac14,10
+OS version: 25D125
+
+*** Sampled system activity (Sat Feb 14 00:42:06 2026 -0500) (503.05ms elapsed) ***
+
+**** GPU usage ****
+
+GPU HW active frequency: 444 MHz
+GPU HW active residency:   0.97% (444 MHz: .97% 612 MHz:   0% 808 MHz:   0% 968 MHz:   0% 1110 MHz:   0% 1236 MHz:   0% 1338 MHz:   0% 1398 MHz:   0%)
+GPU SW requested state: (P1 : 100% P2 :   0% P3 :   0% P4 :   0% P5 :   0% P6 :   0% P7 :   0% P8 :   0%)
+GPU idle residency:  99.03%
+GPU Power: 4 mW
+`
+	gm := &GPUManager{
+		GpuDataMap: make(map[string]*system.GPUData),
+	}
+	valid := gm.parsePowermetricsData([]byte(input))
+	require.True(t, valid)
+
+	g0, ok := gm.GpuDataMap["0"]
+	require.True(t, ok)
+	assert.Equal(t, "Apple GPU", g0.Name)
+	// Usage = 100 - 99.03 = 0.97
+	assert.InDelta(t, 0.97, g0.Usage, 0.01)
+	// 4 mW -> 0.004 W
+	assert.InDelta(t, 0.004, g0.Power, 0.0001)
+	assert.Equal(t, 1.0, g0.Count)
+}
+
+func TestParsePowermetricsDataPartial(t *testing.T) {
+	// Only power line (e.g. older macOS or different sampler output)
+	input := `
+**** GPU usage ****
+GPU Power: 120 mW
+`
+	gm := &GPUManager{
+		GpuDataMap: make(map[string]*system.GPUData),
+	}
+	valid := gm.parsePowermetricsData([]byte(input))
+	require.True(t, valid)
+
+	g0, ok := gm.GpuDataMap["0"]
+	require.True(t, ok)
+	assert.Equal(t, "Apple GPU", g0.Name)
+	assert.InDelta(t, 0.12, g0.Power, 0.001)
+	assert.Equal(t, 1.0, g0.Count)
+}
+
+func TestParseMacmonLine(t *testing.T) {
+	input := `{"all_power":0.6468324661254883,"ane_power":0.0,"cpu_power":0.6359732151031494,"ecpu_usage":[2061,0.1726151406764984],"gpu_power":0.010859241709113121,"gpu_ram_power":0.000965250947047025,"gpu_usage":[503,0.013633215799927711],"memory":{"ram_total":17179869184,"ram_usage":12322914304,"swap_total":0,"swap_usage":0},"pcpu_usage":[1248,0.11792058497667313],"ram_power":0.14885640144348145,"sys_power":10.4955415725708,"temp":{"cpu_temp_avg":23.041261672973633,"gpu_temp_avg":29.44516944885254},"timestamp":"2026-02-17T19:34:27.942556+00:00"}`
+
+	gm := &GPUManager{
+		GpuDataMap: make(map[string]*system.GPUData),
+	}
+	valid := gm.parseMacmonLine([]byte(input))
+	require.True(t, valid)
+
+	g0, ok := gm.GpuDataMap["0"]
+	require.True(t, ok)
+	assert.Equal(t, "Apple GPU", g0.Name)
+	// macmon reports usage fraction 0..1; expect percent conversion.
+	assert.InDelta(t, 1.3633, g0.Usage, 0.05)
+	// power includes gpu_power + gpu_ram_power
+	assert.InDelta(t, 0.011824, g0.Power, 0.0005)
+	assert.InDelta(t, 29.445, g0.Temperature, 0.01)
+	assert.Equal(t, 1.0, g0.Count)
+}

agent/gpu_darwin_unsupported.go

@@ -0,0 +1,9 @@
+//go:build !darwin
+
+package agent
+
+// startPowermetricsCollector is a no-op on non-darwin platforms; the real implementation is in gpu_darwin.go.
+func (gm *GPUManager) startPowermetricsCollector() {}
+
+// startMacmonCollector is a no-op on non-darwin platforms; the real implementation is in gpu_darwin.go.
+func (gm *GPUManager) startMacmonCollector() {}

agent/gpu_intel.go

@@ -7,6 +7,7 @@ import (
 	"strconv"
 	"strings"
 
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"
 )
 
@@ -52,7 +53,7 @@ func (gm *GPUManager) updateIntelFromStats(sample *intelGpuStats) bool {
 func (gm *GPUManager) collectIntelStats() (err error) {
 	// Build command arguments, optionally selecting a device via -d
 	args := []string{"-s", intelGpuStatsInterval, "-l"}
-	if dev, ok := GetEnv("INTEL_GPU_DEVICE"); ok && dev != "" {
+	if dev, ok := utils.GetEnv("INTEL_GPU_DEVICE"); ok && dev != "" {
 		args = append(args, "-d", dev)
 	}
 	cmd := exec.Command(intelGpuStatsCmd, args...)

agent/gpu_nvml_unsupported.go

@@ -13,21 +13,3 @@ func (c *nvmlCollector) init() error {
 }
 
 func (c *nvmlCollector) start() {}
-
-func (c *nvmlCollector) collect() {}
-
-func openLibrary(name string) (uintptr, error) {
-	return 0, fmt.Errorf("nvml not supported on this platform")
-}
-
-func getNVMLPath() string {
-	return ""
-}
-
-func hasSymbol(lib uintptr, symbol string) bool {
-	return false
-}
-
-func (c *nvmlCollector) isGPUActive(bdf string) bool {
-	return true
-}

agent/gpu_nvtop.go

@@ -0,0 +1,160 @@
+package agent
+
+import (
+	"encoding/json"
+	"io"
+	"log/slog"
+	"os/exec"
+	"strconv"
+	"strings"
+	"time"
+
+	"github.com/henrygd/beszel/agent/utils"
+	"github.com/henrygd/beszel/internal/entities/system"
+)
+
+type nvtopSnapshot struct {
+	DeviceName string  `json:"device_name"`
+	Temp       *string `json:"temp"`
+	PowerDraw  *string `json:"power_draw"`
+	GpuUtil    *string `json:"gpu_util"`
+	MemTotal   *string `json:"mem_total"`
+	MemUsed    *string `json:"mem_used"`
+}
+
+// parseNvtopNumber parses nvtop numeric strings with units (C/W/%).
+func parseNvtopNumber(raw string) float64 {
+	cleaned := strings.TrimSpace(raw)
+	cleaned = strings.TrimSuffix(cleaned, "C")
+	cleaned = strings.TrimSuffix(cleaned, "W")
+	cleaned = strings.TrimSuffix(cleaned, "%")
+	val, _ := strconv.ParseFloat(cleaned, 64)
+	return val
+}
+
+// parseNvtopData parses a single nvtop JSON snapshot payload.
+func (gm *GPUManager) parseNvtopData(output []byte) bool {
+	var snapshots []nvtopSnapshot
+	if err := json.Unmarshal(output, &snapshots); err != nil || len(snapshots) == 0 {
+		return false
+	}
+	return gm.updateNvtopSnapshots(snapshots)
+}
+
+// updateNvtopSnapshots applies one decoded nvtop snapshot batch to GPU accumulators.
+func (gm *GPUManager) updateNvtopSnapshots(snapshots []nvtopSnapshot) bool {
+	gm.Lock()
+	defer gm.Unlock()
+
+	valid := false
+	usedIDs := make(map[string]struct{}, len(snapshots))
+	for i, sample := range snapshots {
+		if sample.DeviceName == "" {
+			continue
+		}
+		indexID := "n" + strconv.Itoa(i)
+		id := indexID
+
+		// nvtop ordering can change, so prefer reusing an existing slot with matching device name.
+		if existingByIndex, ok := gm.GpuDataMap[indexID]; ok && existingByIndex.Name != "" && existingByIndex.Name != sample.DeviceName {
+			for existingID, gpu := range gm.GpuDataMap {
+				if !strings.HasPrefix(existingID, "n") {
+					continue
+				}
+				if _, taken := usedIDs[existingID]; taken {
+					continue
+				}
+				if gpu.Name == sample.DeviceName {
+					id = existingID
+					break
+				}
+			}
+		}
+
+		if _, ok := gm.GpuDataMap[id]; !ok {
+			gm.GpuDataMap[id] = &system.GPUData{Name: sample.DeviceName}
+		}
+		gpu := gm.GpuDataMap[id]
+		gpu.Name = sample.DeviceName
+
+		if sample.Temp != nil {
+			gpu.Temperature = parseNvtopNumber(*sample.Temp)
+		}
+		if sample.MemUsed != nil {
+			gpu.MemoryUsed = utils.BytesToMegabytes(parseNvtopNumber(*sample.MemUsed))
+		}
+		if sample.MemTotal != nil {
+			gpu.MemoryTotal = utils.BytesToMegabytes(parseNvtopNumber(*sample.MemTotal))
+		}
+		if sample.GpuUtil != nil {
+			gpu.Usage += parseNvtopNumber(*sample.GpuUtil)
+		}
+		if sample.PowerDraw != nil {
+			gpu.Power += parseNvtopNumber(*sample.PowerDraw)
+		}
+		gpu.Count++
+		usedIDs[id] = struct{}{}
+		valid = true
+	}
+	return valid
+}
+
+// collectNvtopStats runs nvtop loop mode and continuously decodes JSON snapshots.
+func (gm *GPUManager) collectNvtopStats(interval string) error {
+	cmd := exec.Command(nvtopCmd, "-lP", "-d", interval)
+	stdout, err := cmd.StdoutPipe()
+	if err != nil {
+		return err
+	}
+	if err := cmd.Start(); err != nil {
+		return err
+	}
+	defer func() {
+		_ = stdout.Close()
+		if cmd.ProcessState == nil || !cmd.ProcessState.Exited() {
+			_ = cmd.Process.Kill()
+		}
+		_ = cmd.Wait()
+	}()
+
+	decoder := json.NewDecoder(stdout)
+	foundValid := false
+	for {
+		var snapshots []nvtopSnapshot
+		if err := decoder.Decode(&snapshots); err != nil {
+			if err == io.EOF {
+				if foundValid {
+					return nil
+				}
+				return errNoValidData
+			}
+			return err
+		}
+		if gm.updateNvtopSnapshots(snapshots) {
+			foundValid = true
+		}
+	}
+}
+
+// startNvtopCollector starts nvtop collection with retry or fallback callback handling.
+func (gm *GPUManager) startNvtopCollector(interval string, onFailure func()) {
+	go func() {
+		failures := 0
+		for {
+			if err := gm.collectNvtopStats(interval); err != nil {
+				if onFailure != nil {
+					slog.Warn("Error collecting GPU data via nvtop", "err", err)
+					onFailure()
+					return
+				}
+				failures++
+				if failures > maxFailureRetries {
+					break
+				}
+				slog.Warn("Error collecting GPU data via nvtop", "err", err)
+				time.Sleep(retryWaitTime)
+				continue
+			}
+		}
+	}()
+}

agent/gpu_test.go

@@ -1,5 +1,4 @@
 //go:build testing
-// +build testing
 
 package agent
 
@@ -11,6 +10,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"
 
 	"github.com/stretchr/testify/assert"
@@ -250,6 +250,100 @@ func TestParseAmdData(t *testing.T) {
 	}
 }
 
+func TestParseNvtopData(t *testing.T) {
+	input, err := os.ReadFile("test-data/nvtop.json")
+	require.NoError(t, err)
+
+	gm := &GPUManager{
+		GpuDataMap: make(map[string]*system.GPUData),
+	}
+	valid := gm.parseNvtopData(input)
+	require.True(t, valid)
+
+	g0, ok := gm.GpuDataMap["n0"]
+	require.True(t, ok)
+	assert.Equal(t, "NVIDIA GeForce RTX 3050 Ti Laptop GPU", g0.Name)
+	assert.Equal(t, 48.0, g0.Temperature)
+	assert.Equal(t, 5.0, g0.Usage)
+	assert.Equal(t, 13.0, g0.Power)
+	assert.Equal(t, utils.BytesToMegabytes(349372416), g0.MemoryUsed)
+	assert.Equal(t, utils.BytesToMegabytes(4294967296), g0.MemoryTotal)
+	assert.Equal(t, 1.0, g0.Count)
+
+	g1, ok := gm.GpuDataMap["n1"]
+	require.True(t, ok)
+	assert.Equal(t, "AMD Radeon 680M", g1.Name)
+	assert.Equal(t, 48.0, g1.Temperature)
+	assert.Equal(t, 12.0, g1.Usage)
+	assert.Equal(t, 9.0, g1.Power)
+	assert.Equal(t, utils.BytesToMegabytes(1213784064), g1.MemoryUsed)
+	assert.Equal(t, utils.BytesToMegabytes(16929173504), g1.MemoryTotal)
+	assert.Equal(t, 1.0, g1.Count)
+}
+
+func TestUpdateNvtopSnapshotsKeepsDeviceAssociationWhenOrderChanges(t *testing.T) {
+	strPtr := func(s string) *string { return &s }
+
+	gm := &GPUManager{
+		GpuDataMap: make(map[string]*system.GPUData),
+	}
+
+	firstBatch := []nvtopSnapshot{
+		{
+			DeviceName: "NVIDIA GeForce RTX 3050 Ti Laptop GPU",
+			GpuUtil:    strPtr("20%"),
+			PowerDraw:  strPtr("10W"),
+		},
+		{
+			DeviceName: "AMD Radeon 680M",
+			GpuUtil:    strPtr("30%"),
+			PowerDraw:  strPtr("20W"),
+		},
+	}
+	secondBatchSwapped := []nvtopSnapshot{
+		{
+			DeviceName: "AMD Radeon 680M",
+			GpuUtil:    strPtr("40%"),
+			PowerDraw:  strPtr("25W"),
+		},
+		{
+			DeviceName: "NVIDIA GeForce RTX 3050 Ti Laptop GPU",
+			GpuUtil:    strPtr("50%"),
+			PowerDraw:  strPtr("15W"),
+		},
+	}
+
+	require.True(t, gm.updateNvtopSnapshots(firstBatch))
+	require.True(t, gm.updateNvtopSnapshots(secondBatchSwapped))
+
+	nvidia := gm.GpuDataMap["n0"]
+	require.NotNil(t, nvidia)
+	assert.Equal(t, "NVIDIA GeForce RTX 3050 Ti Laptop GPU", nvidia.Name)
+	assert.Equal(t, 70.0, nvidia.Usage)
+	assert.Equal(t, 25.0, nvidia.Power)
+	assert.Equal(t, 2.0, nvidia.Count)
+
+	amd := gm.GpuDataMap["n1"]
+	require.NotNil(t, amd)
+	assert.Equal(t, "AMD Radeon 680M", amd.Name)
+	assert.Equal(t, 70.0, amd.Usage)
+	assert.Equal(t, 45.0, amd.Power)
+	assert.Equal(t, 2.0, amd.Count)
+}
+
+func TestParseCollectorPriority(t *testing.T) {
+	got := parseCollectorPriority(" nvml, nvidia-smi, intel_gpu_top, amd_sysfs, nvtop, rocm-smi, bad ")
+	want := []collectorSource{
+		collectorSourceNVML,
+		collectorSourceNvidiaSMI,
+		collectorSourceIntelGpuTop,
+		collectorSourceAmdSysfs,
+		collectorSourceNVTop,
+		collectorSourceRocmSMI,
+	}
+	assert.Equal(t, want, got)
+}
+
 func TestParseJetsonData(t *testing.T) {
 	tests := []struct {
 		name        string
@@ -987,36 +1081,33 @@ func TestCalculateGPUAverage(t *testing.T) {
 	})
 }
 
-func TestDetectGPUs(t *testing.T) {
+func TestGPUCapabilitiesAndLegacyPriority(t *testing.T) {
 	// Save original PATH
-	origPath := os.Getenv("PATH")
-	defer os.Setenv("PATH", origPath)
-
-	// Set up temp dir with the commands
-	tempDir := t.TempDir()
-	os.Setenv("PATH", tempDir)
+	hasAmdSysfs := (&GPUManager{}).hasAmdSysfs()
 
 	tests := []struct {
 		name           string
-		setupCommands  func() error
+		setupCommands  func(string) error
 		wantNvidiaSmi  bool
 		wantRocmSmi    bool
 		wantTegrastats bool
+		wantNvtop      bool
 		wantErr        bool
 	}{
 		{
 			name: "nvidia-smi not available",
-			setupCommands: func() error {
+			setupCommands: func(_ string) error {
 				return nil
 			},
 			wantNvidiaSmi:  false,
 			wantRocmSmi:    false,
 			wantTegrastats: false,
+			wantNvtop:      false,
 			wantErr:        true,
 		},
 		{
 			name: "nvidia-smi available",
-			setupCommands: func() error {
+			setupCommands: func(tempDir string) error {
 				path := filepath.Join(tempDir, "nvidia-smi")
 				script := `#!/bin/sh
 echo "test"`
@@ -1028,29 +1119,14 @@ echo "test"`
 			wantNvidiaSmi:  true,
 			wantTegrastats: false,
 			wantRocmSmi:    false,
+			wantNvtop:      false,
 			wantErr:        false,
 		},
 		{
 			name: "rocm-smi available",
-			setupCommands: func() error {
+			setupCommands: func(tempDir string) error {
 				path := filepath.Join(tempDir, "rocm-smi")
 				script := `#!/bin/sh
-echo "test"`
-				if err := os.WriteFile(path, []byte(script), 0755); err != nil {
-					return err
-				}
-				return nil
-			},
-			wantNvidiaSmi:  true,
-			wantRocmSmi:    true,
-			wantTegrastats: false,
-			wantErr:        false,
-		},
-		{
-			name: "tegrastats available",
-			setupCommands: func() error {
-				path := filepath.Join(tempDir, "tegrastats")
-				script := `#!/bin/sh
 echo "test"`
 				if err := os.WriteFile(path, []byte(script), 0755); err != nil {
 					return err
@@ -1059,13 +1135,48 @@ echo "test"`
 			},
 			wantNvidiaSmi:  false,
 			wantRocmSmi:    true,
+			wantTegrastats: false,
+			wantNvtop:      false,
+			wantErr:        false,
+		},
+		{
+			name: "tegrastats available",
+			setupCommands: func(tempDir string) error {
+				path := filepath.Join(tempDir, "tegrastats")
+				script := `#!/bin/sh
+echo "test"`
+				if err := os.WriteFile(path, []byte(script), 0755); err != nil {
+					return err
+				}
+				return nil
+			},
+			wantNvidiaSmi:  false,
+			wantRocmSmi:    false,
 			wantTegrastats: true,
+			wantNvtop:      false,
+			wantErr:        false,
+		},
+		{
+			name: "nvtop available",
+			setupCommands: func(tempDir string) error {
+				path := filepath.Join(tempDir, "nvtop")
+				script := `#!/bin/sh
+echo "[]"`
+				if err := os.WriteFile(path, []byte(script), 0755); err != nil {
+					return err
+				}
+				return nil
+			},
+			wantNvidiaSmi:  false,
+			wantRocmSmi:    false,
+			wantTegrastats: false,
+			wantNvtop:      true,
 			wantErr:        false,
 		},
 		{
 			name: "no gpu tools available",
-			setupCommands: func() error {
-				os.Setenv("PATH", "")
+			setupCommands: func(_ string) error {
+				t.Setenv("PATH", "")
 				return nil
 			},
 			wantErr: true,
@@ -1074,36 +1185,56 @@ echo "test"`
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			if err := tt.setupCommands(); err != nil {
+			tempDir := t.TempDir()
+			t.Setenv("PATH", tempDir)
+			if err := tt.setupCommands(tempDir); err != nil {
 				t.Fatal(err)
 			}
 
 			gm := &GPUManager{}
-			err := gm.detectGPUs()
+			caps := gm.discoverGpuCapabilities()
+			var err error
+			if !hasAnyGpuCollector(caps) {
+				err = fmt.Errorf(noGPUFoundMsg)
+			}
+			priorities := gm.resolveLegacyCollectorPriority(caps)
+			hasPriority := func(source collectorSource) bool {
+				for _, s := range priorities {
+					if s == source {
+						return true
+					}
+				}
+				return false
+			}
+			gotNvidiaSmi := hasPriority(collectorSourceNvidiaSMI)
+			gotRocmSmi := hasPriority(collectorSourceRocmSMI)
+			gotTegrastats := caps.hasTegrastats
+			gotNvtop := caps.hasNvtop
 
-			t.Logf("nvidiaSmi: %v, rocmSmi: %v, tegrastats: %v", gm.nvidiaSmi, gm.rocmSmi, gm.tegrastats)
+			t.Logf("nvidiaSmi: %v, rocmSmi: %v, tegrastats: %v", gotNvidiaSmi, gotRocmSmi, gotTegrastats)
 
-			if tt.wantErr {
+			wantErr := tt.wantErr
+			if hasAmdSysfs && (tt.name == "nvidia-smi not available" || tt.name == "no gpu tools available") {
+				wantErr = false
+			}
+			if wantErr {
 				assert.Error(t, err)
 				return
 			}
 
 			assert.NoError(t, err)
-			assert.Equal(t, tt.wantNvidiaSmi, gm.nvidiaSmi)
-			assert.Equal(t, tt.wantRocmSmi, gm.rocmSmi)
-			assert.Equal(t, tt.wantTegrastats, gm.tegrastats)
+			assert.Equal(t, tt.wantNvidiaSmi, gotNvidiaSmi)
+			assert.Equal(t, tt.wantRocmSmi, gotRocmSmi)
+			assert.Equal(t, tt.wantTegrastats, gotTegrastats)
+			assert.Equal(t, tt.wantNvtop, gotNvtop)
 		})
 	}
 }
 
-func TestStartCollector(t *testing.T) {
-	// Save original PATH
-	origPath := os.Getenv("PATH")
-	defer os.Setenv("PATH", origPath)
-
+func TestCollectorStartHelpers(t *testing.T) {
 	// Set up temp dir with the commands
 	dir := t.TempDir()
-	os.Setenv("PATH", dir)
+	t.Setenv("PATH", dir)
 
 	tests := []struct {
 		name     string
@@ -1181,6 +1312,27 @@ echo "11-14-2024 22:54:33 RAM 1024/4096MB GR3D_FREQ 80% tj@70C VDD_GPU_SOC 1000m
 				},
 			},
 		},
+		{
+			name:    "nvtop collector",
+			command: "nvtop",
+			setup: func(t *testing.T) error {
+				path := filepath.Join(dir, "nvtop")
+				script := `#!/bin/sh
+echo '[{"device_name":"NVIDIA Test GPU","temp":"52C","power_draw":"31W","gpu_util":"37%","mem_total":"4294967296","mem_used":"536870912","processes":[]}]'`
+				if err := os.WriteFile(path, []byte(script), 0755); err != nil {
+					return err
+				}
+				return nil
+			},
+			validate: func(t *testing.T, gm *GPUManager) {
+				gpu, exists := gm.GpuDataMap["n0"]
+				assert.True(t, exists)
+				if exists {
+					assert.Equal(t, "NVIDIA Test GPU", gpu.Name)
+					assert.Equal(t, 52.0, gpu.Temperature)
+				}
+			},
+		},
 	}
 
 	for _, tt := range tests {
@@ -1193,13 +1345,161 @@ echo "11-14-2024 22:54:33 RAM 1024/4096MB GR3D_FREQ 80% tj@70C VDD_GPU_SOC 1000m
 					GpuDataMap: make(map[string]*system.GPUData),
 				}
 			}
-			tt.gm.startCollector(tt.command)
+			switch tt.command {
+			case nvidiaSmiCmd:
+				tt.gm.startNvidiaSmiCollector("4")
+			case rocmSmiCmd:
+				tt.gm.startRocmSmiCollector(4300 * time.Millisecond)
+			case tegraStatsCmd:
+				tt.gm.startTegraStatsCollector("3700")
+			case nvtopCmd:
+				tt.gm.startNvtopCollector("30", nil)
+			default:
+				t.Fatalf("unknown test command %q", tt.command)
+			}
 			time.Sleep(50 * time.Millisecond) // Give collector time to run
 			tt.validate(t, tt.gm)
 		})
 	}
 }
 
+func TestNewGPUManagerPriorityNvtopFallback(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("PATH", dir)
+	t.Setenv("BESZEL_AGENT_GPU_COLLECTOR", "nvtop,nvidia-smi")
+
+	nvtopPath := filepath.Join(dir, "nvtop")
+	nvtopScript := `#!/bin/sh
+echo 'not-json'`
+	require.NoError(t, os.WriteFile(nvtopPath, []byte(nvtopScript), 0755))
+
+	nvidiaPath := filepath.Join(dir, "nvidia-smi")
+	nvidiaScript := `#!/bin/sh
+echo "0, NVIDIA Priority GPU, 45, 512, 2048, 12, 25"`
+	require.NoError(t, os.WriteFile(nvidiaPath, []byte(nvidiaScript), 0755))
+
+	gm, err := NewGPUManager()
+	require.NoError(t, err)
+	require.NotNil(t, gm)
+
+	time.Sleep(150 * time.Millisecond)
+	gpu, ok := gm.GpuDataMap["0"]
+	require.True(t, ok)
+	assert.Equal(t, "Priority GPU", gpu.Name)
+	assert.Equal(t, 45.0, gpu.Temperature)
+}
+
+func TestNewGPUManagerPriorityMixedCollectors(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("PATH", dir)
+	t.Setenv("BESZEL_AGENT_GPU_COLLECTOR", "intel_gpu_top,rocm-smi")
+
+	intelPath := filepath.Join(dir, "intel_gpu_top")
+	intelScript := `#!/bin/sh
+echo "Freq MHz      IRQ RC6     Power W     IMC MiB/s             RCS             VCS"
+echo " req  act       /s   %   gpu   pkg     rd     wr       %  se  wa       %  se  wa"
+echo "226  223      338  58  2.00  2.69   1820    965   0.00    0   0    0.00   0   0"
+echo "189  187      412  67  1.80  2.45   1950    823   8.50    2   1    15.00   1   0"
+`
+	require.NoError(t, os.WriteFile(intelPath, []byte(intelScript), 0755))
+
+	rocmPath := filepath.Join(dir, "rocm-smi")
+	rocmScript := `#!/bin/sh
+echo '{"card0": {"Temperature (Sensor edge) (C)": "49.0", "Current Socket Graphics Package Power (W)": "28.159", "GPU use (%)": "0", "VRAM Total Memory (B)": "536870912", "VRAM Total Used Memory (B)": "445550592", "Card Series": "Rembrandt [Radeon 680M]", "GUID": "34756"}}'
+`
+	require.NoError(t, os.WriteFile(rocmPath, []byte(rocmScript), 0755))
+
+	gm, err := NewGPUManager()
+	require.NoError(t, err)
+	require.NotNil(t, gm)
+
+	time.Sleep(150 * time.Millisecond)
+	_, intelOk := gm.GpuDataMap["i0"]
+	_, amdOk := gm.GpuDataMap["34756"]
+	assert.True(t, intelOk)
+	assert.True(t, amdOk)
+}
+
+func TestNewGPUManagerPriorityNvmlFallbackToNvidiaSmi(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("PATH", dir)
+	t.Setenv("BESZEL_AGENT_GPU_COLLECTOR", "nvml,nvidia-smi")
+
+	nvidiaPath := filepath.Join(dir, "nvidia-smi")
+	nvidiaScript := `#!/bin/sh
+echo "0, NVIDIA Fallback GPU, 41, 256, 1024, 8, 14"`
+	require.NoError(t, os.WriteFile(nvidiaPath, []byte(nvidiaScript), 0755))
+
+	gm, err := NewGPUManager()
+	require.NoError(t, err)
+	require.NotNil(t, gm)
+
+	time.Sleep(150 * time.Millisecond)
+	gpu, ok := gm.GpuDataMap["0"]
+	require.True(t, ok)
+	assert.Equal(t, "Fallback GPU", gpu.Name)
+}
+
+func TestNewGPUManagerConfiguredCollectorsMustStart(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("PATH", dir)
+
+	t.Run("configured valid collector unavailable", func(t *testing.T) {
+		t.Setenv("BESZEL_AGENT_GPU_COLLECTOR", "nvidia-smi")
+		gm, err := NewGPUManager()
+		require.Nil(t, gm)
+		require.Error(t, err)
+		assert.Contains(t, err.Error(), "no configured GPU collectors are available")
+	})
+
+	t.Run("configured collector list has only unknown entries", func(t *testing.T) {
+		t.Setenv("BESZEL_AGENT_GPU_COLLECTOR", "bad,unknown")
+		gm, err := NewGPUManager()
+		require.Nil(t, gm)
+		require.Error(t, err)
+		assert.Contains(t, err.Error(), "no configured GPU collectors are available")
+	})
+}
+
+func TestCollectorDefinitionsNvmlDoesNotRequireNvidiaSmi(t *testing.T) {
+	gm := &GPUManager{}
+	definitions := gm.collectorDefinitions(gpuCapabilities{})
+	require.Contains(t, definitions, collectorSourceNVML)
+	assert.True(t, definitions[collectorSourceNVML].available)
+}
+
+func TestNewGPUManagerConfiguredNvmlBypassesCapabilityGate(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("PATH", dir)
+	t.Setenv("BESZEL_AGENT_GPU_COLLECTOR", "nvml")
+
+	gm, err := NewGPUManager()
+	require.Nil(t, gm)
+	require.Error(t, err)
+	assert.Contains(t, err.Error(), "no configured GPU collectors are available")
+	assert.NotContains(t, err.Error(), noGPUFoundMsg)
+}
+
+func TestNewGPUManagerJetsonIgnoresCollectorConfig(t *testing.T) {
+	dir := t.TempDir()
+	t.Setenv("PATH", dir)
+	t.Setenv("BESZEL_AGENT_GPU_COLLECTOR", "nvidia-smi")
+
+	tegraPath := filepath.Join(dir, "tegrastats")
+	tegraScript := `#!/bin/sh
+echo "11-14-2024 22:54:33 RAM 1024/4096MB GR3D_FREQ 80% tj@70C VDD_GPU_SOC 1000mW"`
+	require.NoError(t, os.WriteFile(tegraPath, []byte(tegraScript), 0755))
+
+	gm, err := NewGPUManager()
+	require.NoError(t, err)
+	require.NotNil(t, gm)
+
+	time.Sleep(100 * time.Millisecond)
+	gpu, ok := gm.GpuDataMap["0"]
+	require.True(t, ok)
+	assert.Equal(t, "GPU", gpu.Name)
+}
+
 // TestAccumulationTableDriven tests the accumulation behavior for all three GPU types
 func TestAccumulation(t *testing.T) {
 	type expectedGPUValues struct {
@@ -1417,12 +1717,8 @@ func TestIntelUpdateFromStats(t *testing.T) {
 }
 
 func TestIntelCollectorStreaming(t *testing.T) {
-	// Save and override PATH
-	origPath := os.Getenv("PATH")
-	defer os.Setenv("PATH", origPath)
-
 	dir := t.TempDir()
-	os.Setenv("PATH", dir)
+	t.Setenv("PATH", dir)
 
 	// Create a fake intel_gpu_top that prints -l format with four samples (first will be skipped) and exits
 	scriptPath := filepath.Join(dir, "intel_gpu_top")

agent/handlers_test.go

@@ -1,5 +1,4 @@
 //go:build testing
-// +build testing
 
 package agent
 

agent/health/health_test.go

@@ -1,5 +1,4 @@
 //go:build testing
-// +build testing
 
 package health
 
@@ -37,7 +36,6 @@ func TestHealth(t *testing.T) {
 	})
 
 	// This test uses synctest to simulate time passing.
-	// NOTE: This test requires GOEXPERIMENT=synctest to run.
 	t.Run("check with simulated time", func(t *testing.T) {
 		synctest.Test(t, func(t *testing.T) {
 			// Update the file to set the initial timestamp.

agent/lhm/beszel_lhm.csproj

@@ -8,6 +8,6 @@
   </PropertyGroup>
 
   <ItemGroup>
-    <PackageReference Include="LibreHardwareMonitorLib" Version="0.9.5" />
+    <PackageReference Include="LibreHardwareMonitorLib" Version="0.9.6" />
   </ItemGroup>
 </Project>

agent/mdraid_linux.go

@@ -0,0 +1,233 @@
+//go:build linux
+
+package agent
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+
+	"github.com/henrygd/beszel/agent/utils"
+	"github.com/henrygd/beszel/internal/entities/smart"
+)
+
+// mdraidSysfsRoot is a test hook; production value is "/sys".
+var mdraidSysfsRoot = "/sys"
+
+type mdraidHealth struct {
+	level         string
+	arrayState    string
+	degraded      uint64
+	raidDisks     uint64
+	syncAction    string
+	syncCompleted string
+	syncSpeed     string
+	mismatchCnt   uint64
+	capacity      uint64
+}
+
+// scanMdraidDevices discovers Linux md arrays exposed in sysfs.
+func scanMdraidDevices() []*DeviceInfo {
+	blockDir := filepath.Join(mdraidSysfsRoot, "block")
+	entries, err := os.ReadDir(blockDir)
+	if err != nil {
+		return nil
+	}
+
+	devices := make([]*DeviceInfo, 0, 2)
+	for _, ent := range entries {
+		name := ent.Name()
+		if !isMdraidBlockName(name) {
+			continue
+		}
+		mdDir := filepath.Join(blockDir, name, "md")
+		if !utils.FileExists(filepath.Join(mdDir, "array_state")) {
+			continue
+		}
+
+		devPath := filepath.Join("/dev", name)
+		devices = append(devices, &DeviceInfo{
+			Name:     devPath,
+			Type:     "mdraid",
+			InfoName: devPath + " [mdraid]",
+			Protocol: "MD",
+		})
+	}
+
+	return devices
+}
+
+// collectMdraidHealth reads mdraid health and stores it in SmartDataMap.
+func (sm *SmartManager) collectMdraidHealth(deviceInfo *DeviceInfo) (bool, error) {
+	if deviceInfo == nil || deviceInfo.Name == "" {
+		return false, nil
+	}
+
+	base := filepath.Base(deviceInfo.Name)
+	if !isMdraidBlockName(base) && !strings.EqualFold(deviceInfo.Type, "mdraid") {
+		return false, nil
+	}
+
+	health, ok := readMdraidHealth(base)
+	if !ok {
+		return false, nil
+	}
+
+	deviceInfo.Type = "mdraid"
+	key := fmt.Sprintf("mdraid:%s", base)
+	status := mdraidSmartStatus(health)
+
+	attrs := make([]*smart.SmartAttribute, 0, 10)
+	if health.arrayState != "" {
+		attrs = append(attrs, &smart.SmartAttribute{Name: "ArrayState", RawString: health.arrayState})
+	}
+	if health.level != "" {
+		attrs = append(attrs, &smart.SmartAttribute{Name: "RaidLevel", RawString: health.level})
+	}
+	if health.raidDisks > 0 {
+		attrs = append(attrs, &smart.SmartAttribute{Name: "RaidDisks", RawValue: health.raidDisks})
+	}
+	if health.degraded > 0 {
+		attrs = append(attrs, &smart.SmartAttribute{Name: "Degraded", RawValue: health.degraded})
+	}
+	if health.syncAction != "" {
+		attrs = append(attrs, &smart.SmartAttribute{Name: "SyncAction", RawString: health.syncAction})
+	}
+	if health.syncCompleted != "" {
+		attrs = append(attrs, &smart.SmartAttribute{Name: "SyncCompleted", RawString: health.syncCompleted})
+	}
+	if health.syncSpeed != "" {
+		attrs = append(attrs, &smart.SmartAttribute{Name: "SyncSpeed", RawString: health.syncSpeed})
+	}
+	if health.mismatchCnt > 0 {
+		attrs = append(attrs, &smart.SmartAttribute{Name: "MismatchCount", RawValue: health.mismatchCnt})
+	}
+
+	sm.Lock()
+	defer sm.Unlock()
+
+	if _, exists := sm.SmartDataMap[key]; !exists {
+		sm.SmartDataMap[key] = &smart.SmartData{}
+	}
+
+	data := sm.SmartDataMap[key]
+	data.ModelName = "Linux MD RAID"
+	if health.level != "" {
+		data.ModelName = "Linux MD RAID (" + health.level + ")"
+	}
+	data.Capacity = health.capacity
+	data.SmartStatus = status
+	data.DiskName = filepath.Join("/dev", base)
+	data.DiskType = "mdraid"
+	data.Attributes = attrs
+
+	return true, nil
+}
+
+// readMdraidHealth reads md array health fields from sysfs.
+func readMdraidHealth(blockName string) (mdraidHealth, bool) {
+	var out mdraidHealth
+
+	if !isMdraidBlockName(blockName) {
+		return out, false
+	}
+
+	mdDir := filepath.Join(mdraidSysfsRoot, "block", blockName, "md")
+	arrayState, okState := utils.ReadStringFileOK(filepath.Join(mdDir, "array_state"))
+	if !okState {
+		return out, false
+	}
+
+	out.arrayState = arrayState
+	out.level = utils.ReadStringFile(filepath.Join(mdDir, "level"))
+	out.syncAction = utils.ReadStringFile(filepath.Join(mdDir, "sync_action"))
+	out.syncCompleted = utils.ReadStringFile(filepath.Join(mdDir, "sync_completed"))
+	out.syncSpeed = utils.ReadStringFile(filepath.Join(mdDir, "sync_speed"))
+
+	if val, ok := utils.ReadUintFile(filepath.Join(mdDir, "raid_disks")); ok {
+		out.raidDisks = val
+	}
+	if val, ok := utils.ReadUintFile(filepath.Join(mdDir, "degraded")); ok {
+		out.degraded = val
+	}
+	if val, ok := utils.ReadUintFile(filepath.Join(mdDir, "mismatch_cnt")); ok {
+		out.mismatchCnt = val
+	}
+
+	if capBytes, ok := readMdraidBlockCapacityBytes(blockName, mdraidSysfsRoot); ok {
+		out.capacity = capBytes
+	}
+
+	return out, true
+}
+
+// mdraidSmartStatus maps md state/sync signals to a SMART-like status.
+func mdraidSmartStatus(health mdraidHealth) string {
+	state := strings.ToLower(strings.TrimSpace(health.arrayState))
+	switch state {
+	case "inactive", "faulty", "broken", "stopped":
+		return "FAILED"
+	}
+	// During rebuild/recovery, arrays are often temporarily degraded; report as
+	// warning instead of hard failure while synchronization is in progress.
+	syncAction := strings.ToLower(strings.TrimSpace(health.syncAction))
+	switch syncAction {
+	case "resync", "recover", "reshape":
+		return "WARNING"
+	}
+	if health.degraded > 0 {
+		return "FAILED"
+	}
+	switch syncAction {
+	case "check", "repair":
+		return "WARNING"
+	}
+	switch state {
+	case "clean", "active", "active-idle", "write-pending", "read-auto", "readonly":
+		return "PASSED"
+	}
+	return "UNKNOWN"
+}
+
+// isMdraidBlockName matches /dev/mdN-style block device names.
+func isMdraidBlockName(name string) bool {
+	if !strings.HasPrefix(name, "md") {
+		return false
+	}
+	suffix := strings.TrimPrefix(name, "md")
+	if suffix == "" {
+		return false
+	}
+	for _, c := range suffix {
+		if c < '0' || c > '9' {
+			return false
+		}
+	}
+	return true
+}
+
+// readMdraidBlockCapacityBytes converts block size metadata into bytes.
+func readMdraidBlockCapacityBytes(blockName, root string) (uint64, bool) {
+	sizePath := filepath.Join(root, "block", blockName, "size")
+	lbsPath := filepath.Join(root, "block", blockName, "queue", "logical_block_size")
+
+	sizeStr, ok := utils.ReadStringFileOK(sizePath)
+	if !ok {
+		return 0, false
+	}
+	sectors, err := strconv.ParseUint(sizeStr, 10, 64)
+	if err != nil || sectors == 0 {
+		return 0, false
+	}
+
+	logicalBlockSize := uint64(512)
+	if lbsStr, ok := utils.ReadStringFileOK(lbsPath); ok {
+		if parsed, err := strconv.ParseUint(lbsStr, 10, 64); err == nil && parsed > 0 {
+			logicalBlockSize = parsed
+		}
+	}
+
+	return sectors * logicalBlockSize, true
+}

agent/mdraid_linux_test.go

@@ -0,0 +1,103 @@
+//go:build linux
+
+package agent
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/henrygd/beszel/internal/entities/smart"
+)
+
+func TestMdraidMockSysfsScanAndCollect(t *testing.T) {
+	tmp := t.TempDir()
+	prev := mdraidSysfsRoot
+	mdraidSysfsRoot = tmp
+	t.Cleanup(func() { mdraidSysfsRoot = prev })
+
+	mdDir := filepath.Join(tmp, "block", "md0", "md")
+	queueDir := filepath.Join(tmp, "block", "md0", "queue")
+	if err := os.MkdirAll(mdDir, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.MkdirAll(queueDir, 0o755); err != nil {
+		t.Fatal(err)
+	}
+
+	write := func(path, content string) {
+		t.Helper()
+		if err := os.WriteFile(path, []byte(content), 0o644); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	write(filepath.Join(mdDir, "array_state"), "active\n")
+	write(filepath.Join(mdDir, "level"), "raid1\n")
+	write(filepath.Join(mdDir, "raid_disks"), "2\n")
+	write(filepath.Join(mdDir, "degraded"), "0\n")
+	write(filepath.Join(mdDir, "sync_action"), "resync\n")
+	write(filepath.Join(mdDir, "sync_completed"), "10%\n")
+	write(filepath.Join(mdDir, "sync_speed"), "100M\n")
+	write(filepath.Join(mdDir, "mismatch_cnt"), "0\n")
+	write(filepath.Join(queueDir, "logical_block_size"), "512\n")
+	write(filepath.Join(tmp, "block", "md0", "size"), "2048\n")
+
+	devs := scanMdraidDevices()
+	if len(devs) != 1 {
+		t.Fatalf("scanMdraidDevices() = %d devices, want 1", len(devs))
+	}
+	if devs[0].Name != "/dev/md0" || devs[0].Type != "mdraid" {
+		t.Fatalf("scanMdraidDevices()[0] = %+v, want Name=/dev/md0 Type=mdraid", devs[0])
+	}
+
+	sm := &SmartManager{SmartDataMap: map[string]*smart.SmartData{}}
+	ok, err := sm.collectMdraidHealth(devs[0])
+	if err != nil || !ok {
+		t.Fatalf("collectMdraidHealth() = (ok=%v, err=%v), want (true,nil)", ok, err)
+	}
+	if len(sm.SmartDataMap) != 1 {
+		t.Fatalf("SmartDataMap len=%d, want 1", len(sm.SmartDataMap))
+	}
+	var got *smart.SmartData
+	for _, v := range sm.SmartDataMap {
+		got = v
+		break
+	}
+	if got == nil {
+		t.Fatalf("SmartDataMap value nil")
+	}
+	if got.DiskType != "mdraid" || got.DiskName != "/dev/md0" {
+		t.Fatalf("disk fields = (type=%q name=%q), want (mdraid,/dev/md0)", got.DiskType, got.DiskName)
+	}
+	if got.SmartStatus != "WARNING" {
+		t.Fatalf("SmartStatus=%q, want WARNING", got.SmartStatus)
+	}
+	if got.ModelName == "" || got.Capacity == 0 {
+		t.Fatalf("identity fields = (model=%q cap=%d), want non-empty model and cap>0", got.ModelName, got.Capacity)
+	}
+	if len(got.Attributes) < 5 {
+		t.Fatalf("attributes len=%d, want >= 5", len(got.Attributes))
+	}
+}
+
+func TestMdraidSmartStatus(t *testing.T) {
+	if got := mdraidSmartStatus(mdraidHealth{arrayState: "inactive"}); got != "FAILED" {
+		t.Fatalf("mdraidSmartStatus(inactive) = %q, want FAILED", got)
+	}
+	if got := mdraidSmartStatus(mdraidHealth{arrayState: "active", degraded: 1, syncAction: "recover"}); got != "WARNING" {
+		t.Fatalf("mdraidSmartStatus(degraded+recover) = %q, want WARNING", got)
+	}
+	if got := mdraidSmartStatus(mdraidHealth{arrayState: "active", degraded: 1}); got != "FAILED" {
+		t.Fatalf("mdraidSmartStatus(degraded) = %q, want FAILED", got)
+	}
+	if got := mdraidSmartStatus(mdraidHealth{arrayState: "active", syncAction: "recover"}); got != "WARNING" {
+		t.Fatalf("mdraidSmartStatus(recover) = %q, want WARNING", got)
+	}
+	if got := mdraidSmartStatus(mdraidHealth{arrayState: "clean"}); got != "PASSED" {
+		t.Fatalf("mdraidSmartStatus(clean) = %q, want PASSED", got)
+	}
+	if got := mdraidSmartStatus(mdraidHealth{arrayState: "unknown"}); got != "UNKNOWN" {
+		t.Fatalf("mdraidSmartStatus(unknown) = %q, want UNKNOWN", got)
+	}
+}

agent/mdraid_stub.go

@@ -0,0 +1,11 @@
+//go:build !linux
+
+package agent
+
+func scanMdraidDevices() []*DeviceInfo {
+	return nil
+}
+
+func (sm *SmartManager) collectMdraidHealth(deviceInfo *DeviceInfo) (bool, error) {
+	return false, nil
+}

agent/network.go

@@ -8,6 +8,7 @@ import (
 	"time"
 
 	"github.com/henrygd/beszel/agent/deltatracker"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"
 	psutilNet "github.com/shirou/gopsutil/v4/net"
 )
@@ -94,7 +95,7 @@ func (a *Agent) initializeNetIoStats() {
 	a.netInterfaces = make(map[string]struct{}, 0)
 
 	// parse NICS env var for whitelist / blacklist
-	nicsEnvVal, nicsEnvExists := GetEnv("NICS")
+	nicsEnvVal, nicsEnvExists := utils.GetEnv("NICS")
 	var nicCfg *NicConfig
 	if nicsEnvExists {
 		nicCfg = newNicConfig(nicsEnvVal)
@@ -103,10 +104,7 @@ func (a *Agent) initializeNetIoStats() {
 	// get current network I/O stats and record valid interfaces
 	if netIO, err := psutilNet.IOCounters(true); err == nil {
 		for _, v := range netIO {
-			if nicsEnvExists && !isValidNic(v.Name, nicCfg) {
-				continue
-			}
-			if a.skipNetworkInterface(v) {
+			if skipNetworkInterface(v, nicCfg) {
 				continue
 			}
 			slog.Info("Detected network interface", "name", v.Name, "sent", v.BytesSent, "recv", v.BytesRecv)
@@ -215,10 +213,8 @@ func (a *Agent) applyNetworkTotals(
 	totalBytesSent, totalBytesRecv uint64,
 	bytesSentPerSecond, bytesRecvPerSecond uint64,
 ) {
-	networkSentPs := bytesToMegabytes(float64(bytesSentPerSecond))
-	networkRecvPs := bytesToMegabytes(float64(bytesRecvPerSecond))
-	if networkSentPs > 10_000 || networkRecvPs > 10_000 {
-		slog.Warn("Invalid net stats. Resetting.", "sent", networkSentPs, "recv", networkRecvPs)
+	if bytesSentPerSecond > 10_000_000_000 || bytesRecvPerSecond > 10_000_000_000 {
+		slog.Warn("Invalid net stats. Resetting.", "sent", bytesSentPerSecond, "recv", bytesRecvPerSecond)
 		for _, v := range netIO {
 			if _, exists := a.netInterfaces[v.Name]; !exists {
 				continue
@@ -228,21 +224,29 @@ func (a *Agent) applyNetworkTotals(
 		a.initializeNetIoStats()
 		delete(a.netIoStats, cacheTimeMs)
 		delete(a.netInterfaceDeltaTrackers, cacheTimeMs)
-		systemStats.NetworkSent = 0
-		systemStats.NetworkRecv = 0
 		systemStats.Bandwidth[0], systemStats.Bandwidth[1] = 0, 0
 		return
 	}
 
-	systemStats.NetworkSent = networkSentPs
-	systemStats.NetworkRecv = networkRecvPs
 	systemStats.Bandwidth[0], systemStats.Bandwidth[1] = bytesSentPerSecond, bytesRecvPerSecond
 	nis.BytesSent = totalBytesSent
 	nis.BytesRecv = totalBytesRecv
 	a.netIoStats[cacheTimeMs] = nis
 }
 
-func (a *Agent) skipNetworkInterface(v psutilNet.IOCountersStat) bool {
+// skipNetworkInterface returns true if the network interface should be ignored.
+func skipNetworkInterface(v psutilNet.IOCountersStat, nicCfg *NicConfig) bool {
+	if nicCfg != nil {
+		if !isValidNic(v.Name, nicCfg) {
+			return true
+		}
+		// In whitelist mode, we honor explicit inclusion without auto-filtering.
+		if !nicCfg.isBlacklist {
+			return false
+		}
+		// In blacklist mode, still apply the auto-filter below.
+	}
+
 	switch {
 	case strings.HasPrefix(v.Name, "lo"),
 		strings.HasPrefix(v.Name, "docker"),

agent/network_test.go

@@ -261,6 +261,39 @@ func TestNewNicConfig(t *testing.T) {
 		})
 	}
 }
+func TestSkipNetworkInterface(t *testing.T) {
+	tests := []struct {
+		name       string
+		nic        psutilNet.IOCountersStat
+		nicCfg     *NicConfig
+		expectSkip bool
+	}{
+		{"loopback lo", psutilNet.IOCountersStat{Name: "lo", BytesSent: 100, BytesRecv: 100}, nil, true},
+		{"loopback lo0", psutilNet.IOCountersStat{Name: "lo0", BytesSent: 100, BytesRecv: 100}, nil, true},
+		{"docker prefix", psutilNet.IOCountersStat{Name: "docker0", BytesSent: 100, BytesRecv: 100}, nil, true},
+		{"br- prefix", psutilNet.IOCountersStat{Name: "br-lan", BytesSent: 100, BytesRecv: 100}, nil, true},
+		{"veth prefix", psutilNet.IOCountersStat{Name: "veth0abc", BytesSent: 100, BytesRecv: 100}, nil, true},
+		{"bond prefix", psutilNet.IOCountersStat{Name: "bond0", BytesSent: 100, BytesRecv: 100}, nil, true},
+		{"cali prefix", psutilNet.IOCountersStat{Name: "cali1234", BytesSent: 100, BytesRecv: 100}, nil, true},
+		{"zero BytesRecv", psutilNet.IOCountersStat{Name: "eth0", BytesSent: 100, BytesRecv: 0}, nil, true},
+		{"zero BytesSent", psutilNet.IOCountersStat{Name: "eth0", BytesSent: 0, BytesRecv: 100}, nil, true},
+		{"both zero", psutilNet.IOCountersStat{Name: "eth0", BytesSent: 0, BytesRecv: 0}, nil, true},
+		{"normal eth0", psutilNet.IOCountersStat{Name: "eth0", BytesSent: 100, BytesRecv: 200}, nil, false},
+		{"normal wlan0", psutilNet.IOCountersStat{Name: "wlan0", BytesSent: 1, BytesRecv: 1}, nil, false},
+		{"whitelist overrides skip (docker)", psutilNet.IOCountersStat{Name: "docker0", BytesSent: 100, BytesRecv: 100}, newNicConfig("docker0"), false},
+		{"whitelist overrides skip (lo)", psutilNet.IOCountersStat{Name: "lo", BytesSent: 100, BytesRecv: 100}, newNicConfig("lo"), false},
+		{"whitelist exclusion", psutilNet.IOCountersStat{Name: "eth1", BytesSent: 100, BytesRecv: 100}, newNicConfig("eth0"), true},
+		{"blacklist skip lo", psutilNet.IOCountersStat{Name: "lo", BytesSent: 100, BytesRecv: 100}, newNicConfig("-eth0"), true},
+		{"blacklist explicit eth0", psutilNet.IOCountersStat{Name: "eth0", BytesSent: 100, BytesRecv: 100}, newNicConfig("-eth0"), true},
+		{"blacklist allow eth1", psutilNet.IOCountersStat{Name: "eth1", BytesSent: 100, BytesRecv: 100}, newNicConfig("-eth0"), false},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			assert.Equal(t, tt.expectSkip, skipNetworkInterface(tt.nic, tt.nicCfg))
+		})
+	}
+}
+
 func TestEnsureNetworkInterfacesMap(t *testing.T) {
 	var a Agent
 	var stats system.Stats
@@ -383,8 +416,6 @@ func TestApplyNetworkTotals(t *testing.T) {
 		totalBytesSent        uint64
 		totalBytesRecv        uint64
 		expectReset           bool
-		expectedNetworkSent   float64
-		expectedNetworkRecv   float64
 		expectedBandwidthSent uint64
 		expectedBandwidthRecv uint64
 	}{
@@ -395,8 +426,6 @@ func TestApplyNetworkTotals(t *testing.T) {
 			totalBytesSent:        10000000,
 			totalBytesRecv:        20000000,
 			expectReset:           false,
-			expectedNetworkSent:   0.95, // ~1 MB/s rounded to 2 decimals
-			expectedNetworkRecv:   1.91, // ~2 MB/s rounded to 2 decimals
 			expectedBandwidthSent: 1000000,
 			expectedBandwidthRecv: 2000000,
 		},
@@ -424,18 +453,6 @@ func TestApplyNetworkTotals(t *testing.T) {
 			totalBytesRecv:     20000000,
 			expectReset:        true,
 		},
-		{
-			name:                  "Valid network stats - at threshold boundary",
-			bytesSentPerSecond:    10485750000, // ~9999.99 MB/s (rounds to 9999.99)
-			bytesRecvPerSecond:    10485750000, // ~9999.99 MB/s (rounds to 9999.99)
-			totalBytesSent:        10000000,
-			totalBytesRecv:        20000000,
-			expectReset:           false,
-			expectedNetworkSent:   9999.99,
-			expectedNetworkRecv:   9999.99,
-			expectedBandwidthSent: 10485750000,
-			expectedBandwidthRecv: 10485750000,
-		},
 		{
 			name:                  "Zero values",
 			bytesSentPerSecond:    0,
@@ -443,8 +460,6 @@ func TestApplyNetworkTotals(t *testing.T) {
 			totalBytesSent:        0,
 			totalBytesRecv:        0,
 			expectReset:           false,
-			expectedNetworkSent:   0.0,
-			expectedNetworkRecv:   0.0,
 			expectedBandwidthSent: 0,
 			expectedBandwidthRecv: 0,
 		},
@@ -481,14 +496,10 @@ func TestApplyNetworkTotals(t *testing.T) {
 				// Should have reset network tracking state - maps cleared and stats zeroed
 				assert.NotContains(t, a.netIoStats, cacheTimeMs, "cache entry should be cleared after reset")
 				assert.NotContains(t, a.netInterfaceDeltaTrackers, cacheTimeMs, "tracker should be cleared on reset")
-				assert.Zero(t, systemStats.NetworkSent)
-				assert.Zero(t, systemStats.NetworkRecv)
 				assert.Zero(t, systemStats.Bandwidth[0])
 				assert.Zero(t, systemStats.Bandwidth[1])
 			} else {
 				// Should have applied stats
-				assert.Equal(t, tt.expectedNetworkSent, systemStats.NetworkSent)
-				assert.Equal(t, tt.expectedNetworkRecv, systemStats.NetworkRecv)
 				assert.Equal(t, tt.expectedBandwidthSent, systemStats.Bandwidth[0])
 				assert.Equal(t, tt.expectedBandwidthRecv, systemStats.Bandwidth[1])
 

agent/sensors.go

@@ -2,48 +2,67 @@ package agent
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"log/slog"
 	"path"
 	"runtime"
 	"strconv"
 	"strings"
+	"time"
 	"unicode/utf8"
 
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/system"
 
 	"github.com/shirou/gopsutil/v4/common"
 	"github.com/shirou/gopsutil/v4/sensors"
 )
 
-type SensorConfig struct {
-	context        context.Context
-	sensors        map[string]struct{}
-	primarySensor  string
-	isBlacklist    bool
-	hasWildcards   bool
-	skipCollection bool
-}
-
-func (a *Agent) newSensorConfig() *SensorConfig {
-	primarySensor, _ := GetEnv("PRIMARY_SENSOR")
-	sysSensors, _ := GetEnv("SYS_SENSORS")
-	sensorsEnvVal, sensorsSet := GetEnv("SENSORS")
-	skipCollection := sensorsSet && sensorsEnvVal == ""
-
-	return a.newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal, skipCollection)
-}
+var errTemperatureFetchTimeout = errors.New("temperature collection timed out")
 
 // Matches sensors.TemperaturesWithContext to allow for panic recovery (gopsutil/issues/1832)
 type getTempsFn func(ctx context.Context) ([]sensors.TemperatureStat, error)
 
+type SensorConfig struct {
+	context        context.Context
+	sensors        map[string]struct{}
+	primarySensor  string
+	timeout        time.Duration
+	isBlacklist    bool
+	hasWildcards   bool
+	skipCollection bool
+	firstRun       bool
+}
+
+func (a *Agent) newSensorConfig() *SensorConfig {
+	primarySensor, _ := utils.GetEnv("PRIMARY_SENSOR")
+	sysSensors, _ := utils.GetEnv("SYS_SENSORS")
+	sensorsEnvVal, sensorsSet := utils.GetEnv("SENSORS")
+	skipCollection := sensorsSet && sensorsEnvVal == ""
+	sensorsTimeout, _ := utils.GetEnv("SENSORS_TIMEOUT")
+
+	return a.newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal, sensorsTimeout, skipCollection)
+}
+
 // newSensorConfigWithEnv creates a SensorConfig with the provided environment variables
 // sensorsSet indicates if the SENSORS environment variable was explicitly set (even to empty string)
-func (a *Agent) newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal string, skipCollection bool) *SensorConfig {
+func (a *Agent) newSensorConfigWithEnv(primarySensor, sysSensors, sensorsEnvVal, sensorsTimeout string, skipCollection bool) *SensorConfig {
+	timeout := 2 * time.Second
+	if sensorsTimeout != "" {
+		if d, err := time.ParseDuration(sensorsTimeout); err == nil {
+			timeout = d
+		} else {
+			slog.Warn("Invalid SENSORS_TIMEOUT", "value", sensorsTimeout)
+		}
+	}
+
 	config := &SensorConfig{
 		context:        context.Background(),
 		primarySensor:  primarySensor,
+		timeout:        timeout,
 		skipCollection: skipCollection,
+		firstRun:       true,
 		sensors:        make(map[string]struct{}),
 	}
 
@@ -85,10 +104,12 @@ func (a *Agent) updateTemperatures(systemStats *system.Stats) {
 	// reset high temp
 	a.systemInfo.DashboardTemp = 0
 
-	temps, err := a.getTempsWithPanicRecovery(getSensorTemps)
+	temps, err := a.getTempsWithTimeout(getSensorTemps)
 	if err != nil {
 		// retry once on panic (gopsutil/issues/1832)
-		temps, err = a.getTempsWithPanicRecovery(getSensorTemps)
+		if !errors.Is(err, errTemperatureFetchTimeout) {
+			temps, err = a.getTempsWithTimeout(getSensorTemps)
+		}
 		if err != nil {
 			slog.Warn("Error updating temperatures", "err", err)
 			if len(systemStats.Temperatures) > 0 {
@@ -135,7 +156,7 @@ func (a *Agent) updateTemperatures(systemStats *system.Stats) {
 		case sensorName:
 			a.systemInfo.DashboardTemp = sensor.Temperature
 		}
-		systemStats.Temperatures[sensorName] = twoDecimals(sensor.Temperature)
+		systemStats.Temperatures[sensorName] = utils.TwoDecimals(sensor.Temperature)
 	}
 }
 
@@ -151,6 +172,34 @@ func (a *Agent) getTempsWithPanicRecovery(getTemps getTempsFn) (temps []sensors.
 	return
 }
 
+func (a *Agent) getTempsWithTimeout(getTemps getTempsFn) ([]sensors.TemperatureStat, error) {
+	type result struct {
+		temps []sensors.TemperatureStat
+		err   error
+	}
+
+	// Use a longer timeout on the first run to allow for initialization
+	// (e.g. Windows LHM subprocess startup)
+	timeout := a.sensorConfig.timeout
+	if a.sensorConfig.firstRun {
+		a.sensorConfig.firstRun = false
+		timeout = 10 * time.Second
+	}
+
+	resultCh := make(chan result, 1)
+	go func() {
+		temps, err := a.getTempsWithPanicRecovery(getTemps)
+		resultCh <- result{temps: temps, err: err}
+	}()
+
+	select {
+	case res := <-resultCh:
+		return res.temps, res.err
+	case <-time.After(timeout):
+		return nil, errTemperatureFetchTimeout
+	}
+}
+
 // isValidSensor checks if a sensor is valid based on the sensor name and the sensor config
 func isValidSensor(sensorName string, config *SensorConfig) bool {
 	// if no sensors configured, everything is valid

agent/sensors_test.go

@@ -1,13 +1,12 @@
 //go:build testing
-// +build testing
 
 package agent
 
 import (
 	"context"
 	"fmt"
-	"os"
 	"testing"
+	"time"
 
 	"github.com/henrygd/beszel/internal/entities/system"
 
@@ -169,6 +168,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 		primarySensor  string
 		sysSensors     string
 		sensors        string
+		sensorsTimeout string
 		skipCollection bool
 		expectedConfig *SensorConfig
 	}{
@@ -180,12 +180,37 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			expectedConfig: &SensorConfig{
 				context:        context.Background(),
 				primarySensor:  "",
+				timeout:        2 * time.Second,
 				sensors:        map[string]struct{}{},
 				isBlacklist:    false,
 				hasWildcards:   false,
 				skipCollection: false,
 			},
 		},
+		{
+			name:           "Custom timeout",
+			primarySensor:  "",
+			sysSensors:     "",
+			sensors:        "",
+			sensorsTimeout: "5s",
+			expectedConfig: &SensorConfig{
+				context: context.Background(),
+				timeout: 5 * time.Second,
+				sensors: map[string]struct{}{},
+			},
+		},
+		{
+			name:           "Invalid timeout falls back to default",
+			primarySensor:  "",
+			sysSensors:     "",
+			sensors:        "",
+			sensorsTimeout: "notaduration",
+			expectedConfig: &SensorConfig{
+				context: context.Background(),
+				timeout: 2 * time.Second,
+				sensors: map[string]struct{}{},
+			},
+		},
 		{
 			name:           "Explicitly set to empty string",
 			primarySensor:  "",
@@ -195,6 +220,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			expectedConfig: &SensorConfig{
 				context:        context.Background(),
 				primarySensor:  "",
+				timeout:        2 * time.Second,
 				sensors:        map[string]struct{}{},
 				isBlacklist:    false,
 				hasWildcards:   false,
@@ -209,6 +235,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			expectedConfig: &SensorConfig{
 				context:       context.Background(),
 				primarySensor: "cpu_temp",
+				timeout:       2 * time.Second,
 				sensors:       map[string]struct{}{},
 				isBlacklist:   false,
 				hasWildcards:  false,
@@ -222,6 +249,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			expectedConfig: &SensorConfig{
 				context:       context.Background(),
 				primarySensor: "cpu_temp",
+				timeout:       2 * time.Second,
 				sensors: map[string]struct{}{
 					"cpu_temp": {},
 					"gpu_temp": {},
@@ -238,6 +266,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			expectedConfig: &SensorConfig{
 				context:       context.Background(),
 				primarySensor: "cpu_temp",
+				timeout:       2 * time.Second,
 				sensors: map[string]struct{}{
 					"cpu_temp": {},
 					"gpu_temp": {},
@@ -254,6 +283,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			expectedConfig: &SensorConfig{
 				context:       context.Background(),
 				primarySensor: "cpu_temp",
+				timeout:       2 * time.Second,
 				sensors: map[string]struct{}{
 					"cpu_*":    {},
 					"gpu_temp": {},
@@ -270,6 +300,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			expectedConfig: &SensorConfig{
 				context:       context.Background(),
 				primarySensor: "cpu_temp",
+				timeout:       2 * time.Second,
 				sensors: map[string]struct{}{
 					"cpu_*":    {},
 					"gpu_temp": {},
@@ -285,6 +316,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			sensors:       "cpu_temp",
 			expectedConfig: &SensorConfig{
 				primarySensor: "cpu_temp",
+				timeout:       2 * time.Second,
 				sensors: map[string]struct{}{
 					"cpu_temp": {},
 				},
@@ -296,7 +328,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			result := agent.newSensorConfigWithEnv(tt.primarySensor, tt.sysSensors, tt.sensors, tt.skipCollection)
+			result := agent.newSensorConfigWithEnv(tt.primarySensor, tt.sysSensors, tt.sensors, tt.sensorsTimeout, tt.skipCollection)
 
 			// Check primary sensor
 			assert.Equal(t, tt.expectedConfig.primarySensor, result.primarySensor)
@@ -315,6 +347,7 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 			// Check flags
 			assert.Equal(t, tt.expectedConfig.isBlacklist, result.isBlacklist)
 			assert.Equal(t, tt.expectedConfig.hasWildcards, result.hasWildcards)
+			assert.Equal(t, tt.expectedConfig.timeout, result.timeout)
 
 			// Check context
 			if tt.sysSensors != "" {
@@ -330,40 +363,18 @@ func TestNewSensorConfigWithEnv(t *testing.T) {
 }
 
 func TestNewSensorConfig(t *testing.T) {
-	// Save original environment variables
-	originalPrimary, hasPrimary := os.LookupEnv("BESZEL_AGENT_PRIMARY_SENSOR")
-	originalSys, hasSys := os.LookupEnv("BESZEL_AGENT_SYS_SENSORS")
-	originalSensors, hasSensors := os.LookupEnv("BESZEL_AGENT_SENSORS")
-
-	// Restore environment variables after the test
-	defer func() {
-		// Clean up test environment variables
-		os.Unsetenv("BESZEL_AGENT_PRIMARY_SENSOR")
-		os.Unsetenv("BESZEL_AGENT_SYS_SENSORS")
-		os.Unsetenv("BESZEL_AGENT_SENSORS")
-
-		// Restore original values if they existed
-		if hasPrimary {
-			os.Setenv("BESZEL_AGENT_PRIMARY_SENSOR", originalPrimary)
-		}
-		if hasSys {
-			os.Setenv("BESZEL_AGENT_SYS_SENSORS", originalSys)
-		}
-		if hasSensors {
-			os.Setenv("BESZEL_AGENT_SENSORS", originalSensors)
-		}
-	}()
-
 	// Set test environment variables
-	os.Setenv("BESZEL_AGENT_PRIMARY_SENSOR", "test_primary")
-	os.Setenv("BESZEL_AGENT_SYS_SENSORS", "/test/path")
-	os.Setenv("BESZEL_AGENT_SENSORS", "test_sensor1,test_*,test_sensor3")
+	t.Setenv("BESZEL_AGENT_PRIMARY_SENSOR", "test_primary")
+	t.Setenv("BESZEL_AGENT_SYS_SENSORS", "/test/path")
+	t.Setenv("BESZEL_AGENT_SENSORS", "test_sensor1,test_*,test_sensor3")
+	t.Setenv("BESZEL_AGENT_SENSORS_TIMEOUT", "7s")
 
 	agent := &Agent{}
 	result := agent.newSensorConfig()
 
 	// Verify results
 	assert.Equal(t, "test_primary", result.primarySensor)
+	assert.Equal(t, 7*time.Second, result.timeout)
 	assert.NotNil(t, result.sensors)
 	assert.Equal(t, 3, len(result.sensors))
 	assert.True(t, result.hasWildcards)
@@ -552,3 +563,59 @@ func TestGetTempsWithPanicRecovery(t *testing.T) {
 		})
 	}
 }
+
+func TestGetTempsWithTimeout(t *testing.T) {
+	agent := &Agent{
+		sensorConfig: &SensorConfig{
+			context: context.Background(),
+			timeout: 10 * time.Millisecond,
+		},
+	}
+
+	t.Run("returns temperatures before timeout", func(t *testing.T) {
+		temps, err := agent.getTempsWithTimeout(func(ctx context.Context) ([]sensors.TemperatureStat, error) {
+			return []sensors.TemperatureStat{{SensorKey: "cpu_temp", Temperature: 42}}, nil
+		})
+
+		require.NoError(t, err)
+		require.Len(t, temps, 1)
+		assert.Equal(t, "cpu_temp", temps[0].SensorKey)
+	})
+
+	t.Run("returns timeout error when collector hangs", func(t *testing.T) {
+		temps, err := agent.getTempsWithTimeout(func(ctx context.Context) ([]sensors.TemperatureStat, error) {
+			time.Sleep(50 * time.Millisecond)
+			return []sensors.TemperatureStat{{SensorKey: "cpu_temp", Temperature: 42}}, nil
+		})
+
+		assert.Nil(t, temps)
+		assert.ErrorIs(t, err, errTemperatureFetchTimeout)
+	})
+}
+
+func TestUpdateTemperaturesSkipsOnTimeout(t *testing.T) {
+	agent := &Agent{
+		systemInfo: system.Info{DashboardTemp: 99},
+		sensorConfig: &SensorConfig{
+			context: context.Background(),
+			timeout: 10 * time.Millisecond,
+		},
+	}
+
+	t.Cleanup(func() {
+		getSensorTemps = sensors.TemperaturesWithContext
+	})
+	getSensorTemps = func(ctx context.Context) ([]sensors.TemperatureStat, error) {
+		time.Sleep(50 * time.Millisecond)
+		return nil, nil
+	}
+
+	stats := &system.Stats{
+		Temperatures: map[string]float64{"stale": 50},
+	}
+
+	agent.updateTemperatures(stats)
+
+	assert.Equal(t, 0.0, agent.systemInfo.DashboardTemp)
+	assert.Equal(t, map[string]float64{}, stats.Temperatures)
+}

agent/server.go

@@ -12,6 +12,7 @@ import (
 	"time"
 
 	"github.com/henrygd/beszel"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/common"
 	"github.com/henrygd/beszel/internal/entities/system"
 
@@ -36,7 +37,7 @@ var hubVersions map[string]semver.Version
 // and begins listening for connections. Returns an error if the server
 // is already running or if there's an issue starting the server.
 func (a *Agent) StartServer(opts ServerOptions) error {
-	if disableSSH, _ := GetEnv("DISABLE_SSH"); disableSSH == "true" {
+	if disableSSH, _ := utils.GetEnv("DISABLE_SSH"); disableSSH == "true" {
 		return errors.New("SSH disabled")
 	}
 	if a.server != nil {
@@ -192,7 +193,7 @@ func (a *Agent) handleSSHRequest(w io.Writer, req *common.HubRequest[cbor.RawMes
 
 // handleLegacyStats serves the legacy one-shot stats payload for older hubs
 func (a *Agent) handleLegacyStats(w io.Writer, hubVersion semver.Version) error {
-	stats := a.gatherStats(common.DataRequestOptions{CacheTimeMs: 60_000})
+	stats := a.gatherStats(common.DataRequestOptions{CacheTimeMs: defaultDataCacheTimeMs})
 	return a.writeToSession(w, stats, hubVersion)
 }
 
@@ -238,11 +239,11 @@ func ParseKeys(input string) ([]gossh.PublicKey, error) {
 // and finally defaults to ":45876".
 func GetAddress(addr string) string {
 	if addr == "" {
-		addr, _ = GetEnv("LISTEN")
+		addr, _ = utils.GetEnv("LISTEN")
 	}
 	if addr == "" {
 		// Legacy PORT environment variable support
-		addr, _ = GetEnv("PORT")
+		addr, _ = utils.GetEnv("PORT")
 	}
 	if addr == "" {
 		return ":45876"
@@ -258,7 +259,7 @@ func GetAddress(addr string) string {
 // It checks the NETWORK environment variable first, then infers from
 // the address format: addresses starting with "/" are "unix", others are "tcp".
 func GetNetwork(addr string) string {
-	if network, ok := GetEnv("NETWORK"); ok && network != "" {
+	if network, ok := utils.GetEnv("NETWORK"); ok && network != "" {
 		return network
 	}
 	if strings.HasPrefix(addr, "/") {

agent/server_test.go

@@ -1,5 +1,4 @@
 //go:build testing
-// +build testing
 
 package agent
 
@@ -184,8 +183,7 @@ func TestStartServer(t *testing.T) {
 }
 
 func TestStartServerDisableSSH(t *testing.T) {
-	os.Setenv("BESZEL_AGENT_DISABLE_SSH", "true")
-	defer os.Unsetenv("BESZEL_AGENT_DISABLE_SSH")
+	t.Setenv("BESZEL_AGENT_DISABLE_SSH", "true")
 
 	agent, err := NewAgent("")
 	require.NoError(t, err)

agent/smart.go

@@ -8,6 +8,7 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"log/slog"
 	"os"
 	"os/exec"
 	"path/filepath"
@@ -17,20 +18,22 @@ import (
 	"sync"
 	"time"
 
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/smart"
-
-	"log/slog"
 )
 
 // SmartManager manages data collection for SMART devices
 type SmartManager struct {
 	sync.Mutex
-	SmartDataMap    map[string]*smart.SmartData
-	SmartDevices    []*DeviceInfo
-	refreshMutex    sync.Mutex
-	lastScanTime    time.Time
-	binPath         string
-	excludedDevices map[string]struct{}
+	SmartDataMap       map[string]*smart.SmartData
+	SmartDevices       []*DeviceInfo
+	refreshMutex       sync.Mutex
+	lastScanTime       time.Time
+	smartctlPath       string
+	excludedDevices    map[string]struct{}
+	darwinNvmeOnce     sync.Once
+	darwinNvmeCapacity map[string]uint64      // serial → bytes cache, written once via darwinNvmeOnce
+	darwinNvmeProvider func() ([]byte, error) // overridable for testing
 }
 
 type scanOutput struct {
@@ -157,7 +160,7 @@ func (sm *SmartManager) ScanDevices(force bool) error {
 	currentDevices := sm.devicesSnapshot()
 
 	var configuredDevices []*DeviceInfo
-	if configuredRaw, ok := GetEnv("SMART_DEVICES"); ok {
+	if configuredRaw, ok := utils.GetEnv("SMART_DEVICES"); ok {
 		slog.Info("SMART_DEVICES", "value", configuredRaw)
 		config := strings.TrimSpace(configuredRaw)
 		if config == "" {
@@ -171,27 +174,42 @@ func (sm *SmartManager) ScanDevices(force bool) error {
 		configuredDevices = parsedDevices
 	}
 
-	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
-	defer cancel()
-
-	cmd := exec.CommandContext(ctx, sm.binPath, "--scan", "-j")
-	output, err := cmd.Output()
-
 	var (
 		scanErr        error
 		scannedDevices []*DeviceInfo
 		hasValidScan   bool
 	)
 
-	if err != nil {
-		scanErr = err
-	} else {
-		scannedDevices, hasValidScan = sm.parseScan(output)
-		if !hasValidScan {
-			scanErr = errNoValidSmartData
+	if sm.smartctlPath != "" {
+		ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+		defer cancel()
+
+		cmd := exec.CommandContext(ctx, sm.smartctlPath, "--scan", "-j")
+		output, err := cmd.Output()
+		if err != nil {
+			scanErr = err
+		} else {
+			scannedDevices, hasValidScan = sm.parseScan(output)
+			if !hasValidScan {
+				scanErr = errNoValidSmartData
+			}
 		}
 	}
 
+	// Add eMMC devices (Linux only) by reading sysfs health fields. This does not
+	// require smartctl and does not scan the whole device.
+	if emmcDevices := scanEmmcDevices(); len(emmcDevices) > 0 {
+		scannedDevices = append(scannedDevices, emmcDevices...)
+		hasValidScan = true
+	}
+
+	// Add Linux mdraid arrays by reading sysfs health fields. This does not
+	// require smartctl and does not scan the whole device.
+	if raidDevices := scanMdraidDevices(); len(raidDevices) > 0 {
+		scannedDevices = append(scannedDevices, raidDevices...)
+		hasValidScan = true
+	}
+
 	finalDevices := mergeDeviceLists(currentDevices, scannedDevices, configuredDevices)
 	finalDevices = sm.filterExcludedDevices(finalDevices)
 	sm.updateSmartDevices(finalDevices)
@@ -208,7 +226,7 @@ func (sm *SmartManager) ScanDevices(force bool) error {
 }
 
 func (sm *SmartManager) parseConfiguredDevices(config string) ([]*DeviceInfo, error) {
-	splitChar := os.Getenv("SMART_DEVICES_SEPARATOR")
+	splitChar, _ := utils.GetEnv("SMART_DEVICES_SEPARATOR")
 	if splitChar == "" {
 		splitChar = ","
 	}
@@ -246,7 +264,7 @@ func (sm *SmartManager) parseConfiguredDevices(config string) ([]*DeviceInfo, er
 }
 
 func (sm *SmartManager) refreshExcludedDevices() {
-	rawValue, _ := GetEnv("EXCLUDE_SMART")
+	rawValue, _ := utils.GetEnv("EXCLUDE_SMART")
 	sm.excludedDevices = make(map[string]struct{})
 
 	for entry := range strings.SplitSeq(rawValue, ",") {
@@ -443,6 +461,24 @@ func (sm *SmartManager) CollectSmart(deviceInfo *DeviceInfo) error {
 		return errNoValidSmartData
 	}
 
+	// mdraid health is not exposed via SMART; Linux exposes array state in sysfs.
+	if deviceInfo != nil {
+		if ok, err := sm.collectMdraidHealth(deviceInfo); ok {
+			return err
+		}
+	}
+	// eMMC health is not exposed via SMART on Linux, but the kernel provides
+	// wear / EOL indicators via sysfs. Prefer that path when available.
+	if deviceInfo != nil {
+		if ok, err := sm.collectEmmcHealth(deviceInfo); ok {
+			return err
+		}
+	}
+
+	if sm.smartctlPath == "" {
+		return errNoValidSmartData
+	}
+
 	// slog.Info("collecting SMART data", "device", deviceInfo.Name, "type", deviceInfo.Type, "has_existing_data", sm.hasDataForDevice(deviceInfo.Name))
 
 	// Check if we have any existing data for this device
@@ -453,11 +489,11 @@ func (sm *SmartManager) CollectSmart(deviceInfo *DeviceInfo) error {
 
 	// Try with -n standby first if we have existing data
 	args := sm.smartctlArgs(deviceInfo, hasExistingData)
-	cmd := exec.CommandContext(ctx, sm.binPath, args...)
+	cmd := exec.CommandContext(ctx, sm.smartctlPath, args...)
 	output, err := cmd.CombinedOutput()
 
 	// Check if device is in standby (exit status 2)
-	if exitErr, ok := err.(*exec.ExitError); ok && exitErr.ExitCode() == 2 {
+	if exitErr, ok := errors.AsType[*exec.ExitError](err); ok && exitErr.ExitCode() == 2 {
 		if hasExistingData {
 			// Device is in standby and we have cached data, keep using cache
 			return nil
@@ -466,7 +502,7 @@ func (sm *SmartManager) CollectSmart(deviceInfo *DeviceInfo) error {
 		ctx2, cancel2 := context.WithTimeout(context.Background(), 15*time.Second)
 		defer cancel2()
 		args = sm.smartctlArgs(deviceInfo, false)
-		cmd = exec.CommandContext(ctx2, sm.binPath, args...)
+		cmd = exec.CommandContext(ctx2, sm.smartctlPath, args...)
 		output, err = cmd.CombinedOutput()
 	}
 
@@ -483,7 +519,7 @@ func (sm *SmartManager) CollectSmart(deviceInfo *DeviceInfo) error {
 			ctx3, cancel3 := context.WithTimeout(context.Background(), 15*time.Second)
 			defer cancel3()
 			args = sm.smartctlArgs(deviceInfo, false)
-			cmd = exec.CommandContext(ctx3, sm.binPath, args...)
+			cmd = exec.CommandContext(ctx3, sm.smartctlPath, args...)
 			output, err = cmd.CombinedOutput()
 			hasValidData = sm.parseSmartOutput(deviceInfo, output)
 
@@ -838,15 +874,18 @@ func (sm *SmartManager) parseSmartForSata(output []byte) (bool, int) {
 	smartData.FirmwareVersion = data.FirmwareVersion
 	smartData.Capacity = data.UserCapacity.Bytes
 	smartData.Temperature = data.Temperature.Current
-	if smartData.Temperature == 0 {
-		if temp, ok := temperatureFromAtaDeviceStatistics(data.AtaDeviceStatistics); ok {
-			smartData.Temperature = temp
-		}
-	}
 	smartData.SmartStatus = getSmartStatus(smartData.Temperature, data.SmartStatus.Passed)
 	smartData.DiskName = data.Device.Name
 	smartData.DiskType = data.Device.Type
 
+	// get values from ata_device_statistics if necessary
+	var ataDeviceStats smart.AtaDeviceStatistics
+	if smartData.Temperature == 0 {
+		if temp := findAtaDeviceStatisticsValue(&data, &ataDeviceStats, 5, "Current Temperature", 0, 255); temp != nil {
+			smartData.Temperature = uint8(*temp)
+		}
+	}
+
 	// update SmartAttributes
 	smartData.Attributes = make([]*smart.SmartAttribute, 0, len(data.AtaSmartAttributes.Table))
 	for _, attr := range data.AtaSmartAttributes.Table {
@@ -881,23 +920,20 @@ func getSmartStatus(temperature uint8, passed bool) string {
 	}
 }
 
-func temperatureFromAtaDeviceStatistics(stats smart.AtaDeviceStatistics) (uint8, bool) {
-	entry := findAtaDeviceStatisticsEntry(stats, 5, "Current Temperature")
-	if entry == nil || entry.Value == nil {
-		return 0, false
-	}
-	if *entry.Value > 255 {
-		return 0, false
-	}
-	return uint8(*entry.Value), true
-}
-
 // findAtaDeviceStatisticsEntry centralizes ATA devstat lookups so additional
 // metrics can be pulled from the same structure in the future.
-func findAtaDeviceStatisticsEntry(stats smart.AtaDeviceStatistics, pageNumber uint8, entryName string) *smart.AtaDeviceStatisticsEntry {
-	for pageIdx := range stats.Pages {
-		page := &stats.Pages[pageIdx]
-		if page.Number != pageNumber {
+func findAtaDeviceStatisticsValue(data *smart.SmartInfoForSata, ataDeviceStats *smart.AtaDeviceStatistics, entryNumber uint8, entryName string, minValue, maxValue int64) *int64 {
+	if len(ataDeviceStats.Pages) == 0 {
+		if len(data.AtaDeviceStatistics) == 0 {
+			return nil
+		}
+		if err := json.Unmarshal(data.AtaDeviceStatistics, ataDeviceStats); err != nil {
+			return nil
+		}
+	}
+	for pageIdx := range ataDeviceStats.Pages {
+		page := &ataDeviceStats.Pages[pageIdx]
+		if page.Number != entryNumber {
 			continue
 		}
 		for entryIdx := range page.Table {
@@ -905,7 +941,10 @@ func findAtaDeviceStatisticsEntry(stats smart.AtaDeviceStatistics, pageNumber ui
 			if !strings.EqualFold(entry.Name, entryName) {
 				continue
 			}
-			return entry
+			if entry.Value == nil || *entry.Value < minValue || *entry.Value > maxValue {
+				return nil
+			}
+			return entry.Value
 		}
 	}
 	return nil
@@ -997,6 +1036,52 @@ func parseScsiGigabytesProcessed(value string) int64 {
 	return parsed
 }
 
+// lookupDarwinNvmeCapacity returns the capacity in bytes for a given NVMe serial number on Darwin.
+// It uses system_profiler SPNVMeDataType to get capacity since Apple SSDs don't report user_capacity
+// via smartctl. Results are cached after the first call via sync.Once.
+func (sm *SmartManager) lookupDarwinNvmeCapacity(serial string) uint64 {
+	sm.darwinNvmeOnce.Do(func() {
+		sm.darwinNvmeCapacity = make(map[string]uint64)
+
+		provider := sm.darwinNvmeProvider
+		if provider == nil {
+			provider = func() ([]byte, error) {
+				ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+				defer cancel()
+				return exec.CommandContext(ctx, "system_profiler", "SPNVMeDataType", "-json").Output()
+			}
+		}
+
+		out, err := provider()
+		if err != nil {
+			slog.Debug("system_profiler NVMe lookup failed", "err", err)
+			return
+		}
+
+		var result struct {
+			SPNVMeDataType []struct {
+				Items []struct {
+					DeviceSerial string `json:"device_serial"`
+					SizeInBytes  uint64 `json:"size_in_bytes"`
+				} `json:"_items"`
+			} `json:"SPNVMeDataType"`
+		}
+		if err := json.Unmarshal(out, &result); err != nil {
+			slog.Debug("system_profiler NVMe parse failed", "err", err)
+			return
+		}
+
+		for _, controller := range result.SPNVMeDataType {
+			for _, item := range controller.Items {
+				if item.DeviceSerial != "" && item.SizeInBytes > 0 {
+					sm.darwinNvmeCapacity[item.DeviceSerial] = item.SizeInBytes
+				}
+			}
+		}
+	})
+	return sm.darwinNvmeCapacity[serial]
+}
+
 // parseSmartForNvme parses the output of smartctl --all -j /dev/nvmeX and updates the SmartDataMap
 // Returns hasValidData and exitStatus
 func (sm *SmartManager) parseSmartForNvme(output []byte) (bool, int) {
@@ -1033,6 +1118,12 @@ func (sm *SmartManager) parseSmartForNvme(output []byte) (bool, int) {
 	smartData.SerialNumber = data.SerialNumber
 	smartData.FirmwareVersion = data.FirmwareVersion
 	smartData.Capacity = data.UserCapacity.Bytes
+	if smartData.Capacity == 0 {
+		smartData.Capacity = data.NVMeTotalCapacity
+	}
+	if smartData.Capacity == 0 && (runtime.GOOS == "darwin" || sm.darwinNvmeProvider != nil) {
+		smartData.Capacity = sm.lookupDarwinNvmeCapacity(data.SerialNumber)
+	}
 	smartData.Temperature = data.NVMeSmartHealthInformationLog.Temperature
 	smartData.SmartStatus = getSmartStatus(smartData.Temperature, data.SmartStatus.Passed)
 	smartData.DiskName = data.Device.Name
@@ -1068,32 +1159,21 @@ func (sm *SmartManager) parseSmartForNvme(output []byte) (bool, int) {
 
 // detectSmartctl checks if smartctl is installed, returns an error if not
 func (sm *SmartManager) detectSmartctl() (string, error) {
-	isWindows := runtime.GOOS == "windows"
-
-	// Load embedded smartctl.exe for Windows amd64 builds.
-	if isWindows && runtime.GOARCH == "amd64" {
-		if path, err := ensureEmbeddedSmartctl(); err == nil {
-			return path, nil
+	if runtime.GOOS == "windows" {
+		// Load embedded smartctl.exe for Windows amd64 builds.
+		if runtime.GOARCH == "amd64" {
+			if path, err := ensureEmbeddedSmartctl(); err == nil {
+				return path, nil
+			}
 		}
-	}
-
-	if path, err := exec.LookPath("smartctl"); err == nil {
-		return path, nil
-	}
-	locations := []string{}
-	if isWindows {
-		locations = append(locations,
-			"C:\\Program Files\\smartmontools\\bin\\smartctl.exe",
-		)
-	} else {
-		locations = append(locations, "/opt/homebrew/bin/smartctl")
-	}
-	for _, location := range locations {
+		// Try to find smartctl in the default installation location
+		const location = "C:\\Program Files\\smartmontools\\bin\\smartctl.exe"
 		if _, err := os.Stat(location); err == nil {
 			return location, nil
 		}
 	}
-	return "", errors.New("smartctl not found")
+
+	return utils.LookPathHomebrew("smartctl")
 }
 
 // isNvmeControllerPath checks if the path matches an NVMe controller pattern
@@ -1124,11 +1204,17 @@ func NewSmartManager() (*SmartManager, error) {
 	}
 	sm.refreshExcludedDevices()
 	path, err := sm.detectSmartctl()
+	slog.Debug("smartctl", "path", path, "err", err)
 	if err != nil {
-		slog.Debug(err.Error())
+		// Keep the previous fail-fast behavior unless this Linux host exposes
+		// eMMC or mdraid health via sysfs, in which case smartctl is optional.
+		if runtime.GOOS == "linux" {
+			if len(scanEmmcDevices()) > 0 || len(scanMdraidDevices()) > 0 {
+				return sm, nil
+			}
+		}
 		return nil, err
 	}
-	slog.Debug("smartctl", "path", path)
-	sm.binPath = path
+	sm.smartctlPath = path
 	return sm, nil
 }

agent/smart_test.go

@@ -1,5 +1,4 @@
 //go:build testing
-// +build testing
 
 package agent
 
@@ -122,6 +121,78 @@ func TestParseSmartForSataDeviceStatisticsTemperature(t *testing.T) {
 	assert.Equal(t, uint8(22), deviceData.Temperature)
 }
 
+func TestParseSmartForSataAtaDeviceStatistics(t *testing.T) {
+	// tests that ata_device_statistics values are parsed correctly
+	jsonPayload := []byte(`{
+		"smartctl": {"exit_status": 0},
+		"device": {"name": "/dev/sdb", "type": "sat"},
+		"model_name": "SanDisk SSD U110 16GB",
+		"serial_number": "lksjfh23lhj",
+		"firmware_version": "U21B001",
+		"user_capacity": {"bytes": 16013942784},
+		"smart_status": {"passed": true},
+		"ata_smart_attributes": {"table": []},
+		"ata_device_statistics": {
+			"pages": [
+				{
+					"number": 5,
+					"name": "Temperature Statistics",
+					"table": [
+						{"name": "Current Temperature", "value": 43, "flags": {"valid": true}},
+						{"name": "Specified Minimum Operating Temperature", "value": -20, "flags": {"valid": true}}
+					]
+				}
+			]
+		}
+	}`)
+
+	sm := &SmartManager{SmartDataMap: make(map[string]*smart.SmartData)}
+	hasData, exitStatus := sm.parseSmartForSata(jsonPayload)
+	require.True(t, hasData)
+	assert.Equal(t, 0, exitStatus)
+
+	deviceData, ok := sm.SmartDataMap["lksjfh23lhj"]
+	require.True(t, ok, "expected smart data entry for serial lksjfh23lhj")
+	assert.Equal(t, uint8(43), deviceData.Temperature)
+}
+
+func TestParseSmartForSataNegativeDeviceStatistics(t *testing.T) {
+	// Tests that negative values in ata_device_statistics (e.g. min operating temp)
+	// do not cause the entire SAT parser to fail.
+	jsonPayload := []byte(`{
+		"smartctl": {"exit_status": 0},
+		"device": {"name": "/dev/sdb", "type": "sat"},
+		"model_name": "SanDisk SSD U110 16GB",
+		"serial_number": "NEGATIVE123",
+		"firmware_version": "U21B001",
+		"user_capacity": {"bytes": 16013942784},
+		"smart_status": {"passed": true},
+		"temperature": {"current": 38},
+		"ata_smart_attributes": {"table": []},
+		"ata_device_statistics": {
+			"pages": [
+				{
+					"number": 5,
+					"name": "Temperature Statistics",
+					"table": [
+						{"name": "Current Temperature", "value": 38, "flags": {"valid": true}},
+						{"name": "Specified Minimum Operating Temperature", "value": -20, "flags": {"valid": true}}
+					]
+				}
+			]
+		}
+	}`)
+
+	sm := &SmartManager{SmartDataMap: make(map[string]*smart.SmartData)}
+	hasData, exitStatus := sm.parseSmartForSata(jsonPayload)
+	require.True(t, hasData)
+	assert.Equal(t, 0, exitStatus)
+
+	deviceData, ok := sm.SmartDataMap["NEGATIVE123"]
+	require.True(t, ok, "expected smart data entry for serial NEGATIVE123")
+	assert.Equal(t, uint8(38), deviceData.Temperature)
+}
+
 func TestParseSmartForSataParentheticalRawValue(t *testing.T) {
 	jsonPayload := []byte(`{
 		"smartctl": {"exit_status": 0},
@@ -728,6 +799,182 @@ func TestIsVirtualDeviceScsi(t *testing.T) {
 	}
 }
 
+func TestFindAtaDeviceStatisticsValue(t *testing.T) {
+	val42 := int64(42)
+	val100 := int64(100)
+	valMinus20 := int64(-20)
+
+	tests := []struct {
+		name           string
+		data           smart.SmartInfoForSata
+		ataDeviceStats smart.AtaDeviceStatistics
+		entryNumber    uint8
+		entryName      string
+		minValue       int64
+		maxValue       int64
+		expectedValue  *int64
+	}{
+		{
+			name: "value in ataDeviceStats",
+			ataDeviceStats: smart.AtaDeviceStatistics{
+				Pages: []smart.AtaDeviceStatisticsPage{
+					{
+						Number: 5,
+						Table: []smart.AtaDeviceStatisticsEntry{
+							{Name: "Current Temperature", Value: &val42},
+						},
+					},
+				},
+			},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      100,
+			expectedValue: &val42,
+		},
+		{
+			name: "value unmarshaled from data",
+			data: smart.SmartInfoForSata{
+				AtaDeviceStatistics: []byte(`{"pages":[{"number":5,"table":[{"name":"Current Temperature","value":100}]}]}`),
+			},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      255,
+			expectedValue: &val100,
+		},
+		{
+			name: "value out of range (too high)",
+			ataDeviceStats: smart.AtaDeviceStatistics{
+				Pages: []smart.AtaDeviceStatisticsPage{
+					{
+						Number: 5,
+						Table: []smart.AtaDeviceStatisticsEntry{
+							{Name: "Current Temperature", Value: &val100},
+						},
+					},
+				},
+			},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      50,
+			expectedValue: nil,
+		},
+		{
+			name: "value out of range (too low)",
+			ataDeviceStats: smart.AtaDeviceStatistics{
+				Pages: []smart.AtaDeviceStatisticsPage{
+					{
+						Number: 5,
+						Table: []smart.AtaDeviceStatisticsEntry{
+							{Name: "Min Temp", Value: &valMinus20},
+						},
+					},
+				},
+			},
+			entryNumber:   5,
+			entryName:     "Min Temp",
+			minValue:      0,
+			maxValue:      100,
+			expectedValue: nil,
+		},
+		{
+			name:          "no statistics available",
+			data:          smart.SmartInfoForSata{},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      255,
+			expectedValue: nil,
+		},
+		{
+			name: "wrong page number",
+			ataDeviceStats: smart.AtaDeviceStatistics{
+				Pages: []smart.AtaDeviceStatisticsPage{
+					{
+						Number: 1,
+						Table: []smart.AtaDeviceStatisticsEntry{
+							{Name: "Current Temperature", Value: &val42},
+						},
+					},
+				},
+			},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      100,
+			expectedValue: nil,
+		},
+		{
+			name: "wrong entry name",
+			ataDeviceStats: smart.AtaDeviceStatistics{
+				Pages: []smart.AtaDeviceStatisticsPage{
+					{
+						Number: 5,
+						Table: []smart.AtaDeviceStatisticsEntry{
+							{Name: "Other Stat", Value: &val42},
+						},
+					},
+				},
+			},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      100,
+			expectedValue: nil,
+		},
+		{
+			name: "case insensitive name match",
+			ataDeviceStats: smart.AtaDeviceStatistics{
+				Pages: []smart.AtaDeviceStatisticsPage{
+					{
+						Number: 5,
+						Table: []smart.AtaDeviceStatisticsEntry{
+							{Name: "CURRENT TEMPERATURE", Value: &val42},
+						},
+					},
+				},
+			},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      100,
+			expectedValue: &val42,
+		},
+		{
+			name: "entry value is nil",
+			ataDeviceStats: smart.AtaDeviceStatistics{
+				Pages: []smart.AtaDeviceStatisticsPage{
+					{
+						Number: 5,
+						Table: []smart.AtaDeviceStatisticsEntry{
+							{Name: "Current Temperature", Value: nil},
+						},
+					},
+				},
+			},
+			entryNumber:   5,
+			entryName:     "Current Temperature",
+			minValue:      0,
+			maxValue:      100,
+			expectedValue: nil,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := findAtaDeviceStatisticsValue(&tt.data, &tt.ataDeviceStats, tt.entryNumber, tt.entryName, tt.minValue, tt.maxValue)
+			if tt.expectedValue == nil {
+				assert.Nil(t, result)
+			} else {
+				require.NotNil(t, result)
+				assert.Equal(t, *tt.expectedValue, *result)
+			}
+		})
+	}
+}
+
 func TestRefreshExcludedDevices(t *testing.T) {
 	tests := []struct {
 		name         string
@@ -788,7 +1035,7 @@ func TestRefreshExcludedDevices(t *testing.T) {
 				t.Setenv("EXCLUDE_SMART", tt.envValue)
 			} else {
 				// Ensure env var is not set for empty test
-				os.Unsetenv("EXCLUDE_SMART")
+				t.Setenv("EXCLUDE_SMART", "")
 			}
 
 			sm := &SmartManager{}
@@ -952,3 +1199,81 @@ func TestIsNvmeControllerPath(t *testing.T) {
 		})
 	}
 }
+
+func TestParseSmartForNvmeAppleSSD(t *testing.T) {
+	// Apple SSDs don't report user_capacity via smartctl; capacity should be fetched
+	// from system_profiler via the darwinNvmeProvider fallback.
+	fixturePath := filepath.Join("test-data", "smart", "apple_nvme.json")
+	data, err := os.ReadFile(fixturePath)
+	require.NoError(t, err)
+
+	providerCalls := 0
+	fakeProvider := func() ([]byte, error) {
+		providerCalls++
+		return []byte(`{
+			"SPNVMeDataType": [{
+				"_items": [{
+					"device_serial": "0ba0147940253c15",
+					"size_in_bytes": 251000193024
+				}]
+			}]
+		}`), nil
+	}
+
+	sm := &SmartManager{
+		SmartDataMap:       make(map[string]*smart.SmartData),
+		darwinNvmeProvider: fakeProvider,
+	}
+
+	hasData, _ := sm.parseSmartForNvme(data)
+	require.True(t, hasData)
+
+	deviceData, ok := sm.SmartDataMap["0ba0147940253c15"]
+	require.True(t, ok)
+	assert.Equal(t, "APPLE SSD AP0256Q", deviceData.ModelName)
+	assert.Equal(t, uint64(251000193024), deviceData.Capacity)
+	assert.Equal(t, uint8(42), deviceData.Temperature)
+	assert.Equal(t, "PASSED", deviceData.SmartStatus)
+	assert.Equal(t, 1, providerCalls, "system_profiler should be called once")
+
+	// Second parse: provider should NOT be called again (cache hit)
+	_, _ = sm.parseSmartForNvme(data)
+	assert.Equal(t, 1, providerCalls, "system_profiler should not be called again after caching")
+}
+
+func TestLookupDarwinNvmeCapacityMultipleDisks(t *testing.T) {
+	fakeProvider := func() ([]byte, error) {
+		return []byte(`{
+			"SPNVMeDataType": [
+				{
+					"_items": [
+						{"device_serial": "serial-disk0", "size_in_bytes": 251000193024},
+						{"device_serial": "serial-disk1", "size_in_bytes": 1000204886016}
+					]
+				},
+				{
+					"_items": [
+						{"device_serial": "serial-disk2", "size_in_bytes": 512110190592}
+					]
+				}
+			]
+		}`), nil
+	}
+
+	sm := &SmartManager{darwinNvmeProvider: fakeProvider}
+	assert.Equal(t, uint64(251000193024), sm.lookupDarwinNvmeCapacity("serial-disk0"))
+	assert.Equal(t, uint64(1000204886016), sm.lookupDarwinNvmeCapacity("serial-disk1"))
+	assert.Equal(t, uint64(512110190592), sm.lookupDarwinNvmeCapacity("serial-disk2"))
+	assert.Equal(t, uint64(0), sm.lookupDarwinNvmeCapacity("unknown-serial"))
+}
+
+func TestLookupDarwinNvmeCapacityProviderError(t *testing.T) {
+	fakeProvider := func() ([]byte, error) {
+		return nil, errors.New("system_profiler not found")
+	}
+
+	sm := &SmartManager{darwinNvmeProvider: fakeProvider}
+	assert.Equal(t, uint64(0), sm.lookupDarwinNvmeCapacity("any-serial"))
+	// Cache should be initialized even on error so we don't retry (Once already fired)
+	assert.NotNil(t, sm.darwinNvmeCapacity)
+}

agent/system.go

@@ -7,12 +7,12 @@ import (
 	"log/slog"
 	"os"
 	"runtime"
-	"strconv"
 	"strings"
-	"time"
 
 	"github.com/henrygd/beszel"
 	"github.com/henrygd/beszel/agent/battery"
+	"github.com/henrygd/beszel/agent/utils"
+	"github.com/henrygd/beszel/agent/zfs"
 	"github.com/henrygd/beszel/internal/entities/container"
 	"github.com/henrygd/beszel/internal/entities/system"
 
@@ -22,13 +22,6 @@ import (
 	"github.com/shirou/gopsutil/v4/mem"
 )
 
-// prevDisk stores previous per-device disk counters for a given cache interval
-type prevDisk struct {
-	readBytes  uint64
-	writeBytes uint64
-	at         time.Time
-}
-
 // Sets initial / non-changing values about the host system
 func (a *Agent) refreshSystemDetails() {
 	a.systemInfo.AgentVersion = beszel.Version
@@ -107,13 +100,33 @@ func (a *Agent) refreshSystemDetails() {
 	}
 
 	// zfs
-	if _, err := getARCSize(); err != nil {
+	if _, err := zfs.ARCSize(); err != nil {
 		slog.Debug("Not monitoring ZFS ARC", "err", err)
 	} else {
 		a.zfs = true
 	}
 }
 
+// attachSystemDetails returns details only for fresh default-interval responses.
+func (a *Agent) attachSystemDetails(data *system.CombinedData, cacheTimeMs uint16, includeRequested bool) *system.CombinedData {
+	if cacheTimeMs != defaultDataCacheTimeMs || (!includeRequested && !a.detailsDirty) {
+		return data
+	}
+
+	// copy data to avoid adding details to the original cached struct
+	response := *data
+	response.Details = &a.systemDetails
+	a.detailsDirty = false
+	return &response
+}
+
+// updateSystemDetails applies a mutation to the static details payload and marks
+// it for inclusion on the next fresh default-interval response.
+func (a *Agent) updateSystemDetails(updateFunc func(details *system.Details)) {
+	updateFunc(&a.systemDetails)
+	a.detailsDirty = true
+}
+
 // Returns current info, stats about the host system
 func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 	var systemStats system.Stats
@@ -127,13 +140,13 @@ func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 	// cpu metrics
 	cpuMetrics, err := getCpuMetrics(cacheTimeMs)
 	if err == nil {
-		systemStats.Cpu = twoDecimals(cpuMetrics.Total)
+		systemStats.Cpu = utils.TwoDecimals(cpuMetrics.Total)
 		systemStats.CpuBreakdown = []float64{
-			twoDecimals(cpuMetrics.User),
-			twoDecimals(cpuMetrics.System),
-			twoDecimals(cpuMetrics.Iowait),
-			twoDecimals(cpuMetrics.Steal),
-			twoDecimals(cpuMetrics.Idle),
+			utils.TwoDecimals(cpuMetrics.User),
+			utils.TwoDecimals(cpuMetrics.System),
+			utils.TwoDecimals(cpuMetrics.Iowait),
+			utils.TwoDecimals(cpuMetrics.Steal),
+			utils.TwoDecimals(cpuMetrics.Idle),
 		}
 	} else {
 		slog.Error("Error getting cpu metrics", "err", err)
@@ -157,8 +170,8 @@ func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 	// memory
 	if v, err := mem.VirtualMemory(); err == nil {
 		// swap
-		systemStats.Swap = bytesToGigabytes(v.SwapTotal)
-		systemStats.SwapUsed = bytesToGigabytes(v.SwapTotal - v.SwapFree - v.SwapCached)
+		systemStats.Swap = utils.BytesToGigabytes(v.SwapTotal)
+		systemStats.SwapUsed = utils.BytesToGigabytes(v.SwapTotal - v.SwapFree - v.SwapCached)
 		// cache + buffers value for default mem calculation
 		// note: gopsutil automatically adds SReclaimable to v.Cached
 		cacheBuff := v.Cached + v.Buffers - v.Shared
@@ -178,16 +191,16 @@ func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 		// }
 		// subtract ZFS ARC size from used memory and add as its own category
 		if a.zfs {
-			if arcSize, _ := getARCSize(); arcSize > 0 && arcSize < v.Used {
+			if arcSize, _ := zfs.ARCSize(); arcSize > 0 && arcSize < v.Used {
 				v.Used = v.Used - arcSize
 				v.UsedPercent = float64(v.Used) / float64(v.Total) * 100.0
-				systemStats.MemZfsArc = bytesToGigabytes(arcSize)
+				systemStats.MemZfsArc = utils.BytesToGigabytes(arcSize)
 			}
 		}
-		systemStats.Mem = bytesToGigabytes(v.Total)
-		systemStats.MemBuffCache = bytesToGigabytes(cacheBuff)
-		systemStats.MemUsed = bytesToGigabytes(v.Used)
-		systemStats.MemPct = twoDecimals(v.UsedPercent)
+		systemStats.Mem = utils.BytesToGigabytes(v.Total)
+		systemStats.MemBuffCache = utils.BytesToGigabytes(cacheBuff)
+		systemStats.MemUsed = utils.BytesToGigabytes(v.Used)
+		systemStats.MemPct = utils.TwoDecimals(v.UsedPercent)
 	}
 
 	// disk usage
@@ -250,32 +263,6 @@ func (a *Agent) getSystemStats(cacheTimeMs uint16) system.Stats {
 	return systemStats
 }
 
-// Returns the size of the ZFS ARC memory cache in bytes
-func getARCSize() (uint64, error) {
-	file, err := os.Open("/proc/spl/kstat/zfs/arcstats")
-	if err != nil {
-		return 0, err
-	}
-	defer file.Close()
-
-	// Scan the lines
-	scanner := bufio.NewScanner(file)
-	for scanner.Scan() {
-		line := scanner.Text()
-		if strings.HasPrefix(line, "size") {
-			// Example line: size 4 15032385536
-			fields := strings.Fields(line)
-			if len(fields) < 3 {
-				return 0, err
-			}
-			// Return the size as uint64
-			return strconv.ParseUint(fields[2], 10, 64)
-		}
-	}
-
-	return 0, fmt.Errorf("failed to parse size field")
-}
-
 // getOsPrettyName attempts to get the pretty OS name from /etc/os-release on Linux systems
 func getOsPrettyName() (string, error) {
 	file, err := os.Open("/etc/os-release")

agent/system_test.go

@@ -0,0 +1,61 @@
+package agent
+
+import (
+	"testing"
+
+	"github.com/henrygd/beszel/internal/common"
+	"github.com/henrygd/beszel/internal/entities/system"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestGatherStatsDoesNotAttachDetailsToCachedRequests(t *testing.T) {
+	agent := &Agent{
+		cache:         NewSystemDataCache(),
+		systemDetails: system.Details{Hostname: "updated-host", Podman: true},
+		detailsDirty:  true,
+	}
+	cached := &system.CombinedData{
+		Info: system.Info{Hostname: "cached-host"},
+	}
+	agent.cache.Set(cached, defaultDataCacheTimeMs)
+
+	response := agent.gatherStats(common.DataRequestOptions{CacheTimeMs: defaultDataCacheTimeMs})
+
+	assert.Same(t, cached, response)
+	assert.Nil(t, response.Details)
+	assert.True(t, agent.detailsDirty)
+	assert.Equal(t, "cached-host", response.Info.Hostname)
+	assert.Nil(t, cached.Details)
+
+	secondResponse := agent.gatherStats(common.DataRequestOptions{CacheTimeMs: defaultDataCacheTimeMs})
+	assert.Same(t, cached, secondResponse)
+	assert.Nil(t, secondResponse.Details)
+}
+
+func TestUpdateSystemDetailsMarksDetailsDirty(t *testing.T) {
+	agent := &Agent{}
+
+	agent.updateSystemDetails(func(details *system.Details) {
+		details.Hostname = "updated-host"
+		details.Podman = true
+	})
+
+	assert.True(t, agent.detailsDirty)
+	assert.Equal(t, "updated-host", agent.systemDetails.Hostname)
+	assert.True(t, agent.systemDetails.Podman)
+
+	original := &system.CombinedData{}
+	realTimeResponse := agent.attachSystemDetails(original, 1000, true)
+	assert.Same(t, original, realTimeResponse)
+	assert.Nil(t, realTimeResponse.Details)
+	assert.True(t, agent.detailsDirty)
+
+	response := agent.attachSystemDetails(original, defaultDataCacheTimeMs, false)
+	require.NotNil(t, response.Details)
+	assert.NotSame(t, original, response)
+	assert.Equal(t, "updated-host", response.Details.Hostname)
+	assert.True(t, response.Details.Podman)
+	assert.False(t, agent.detailsDirty)
+	assert.Nil(t, original.Details)
+}

agent/systemd.go

@@ -15,6 +15,7 @@ import (
 	"time"
 
 	"github.com/coreos/go-systemd/v22/dbus"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/henrygd/beszel/internal/entities/systemd"
 )
 
@@ -49,7 +50,7 @@ func isSystemdAvailable() bool {
 
 // newSystemdManager creates a new systemdManager.
 func newSystemdManager() (*systemdManager, error) {
-	if skipSystemd, _ := GetEnv("SKIP_SYSTEMD"); skipSystemd == "true" {
+	if skipSystemd, _ := utils.GetEnv("SKIP_SYSTEMD"); skipSystemd == "true" {
 		return nil, nil
 	}
 
@@ -294,13 +295,13 @@ func unescapeServiceName(name string) string {
 // otherwise defaults to "*service".
 func getServicePatterns() []string {
 	patterns := []string{}
-	if envPatterns, _ := GetEnv("SERVICE_PATTERNS"); envPatterns != "" {
+	if envPatterns, _ := utils.GetEnv("SERVICE_PATTERNS"); envPatterns != "" {
 		for pattern := range strings.SplitSeq(envPatterns, ",") {
 			pattern = strings.TrimSpace(pattern)
 			if pattern == "" {
 				continue
 			}
-			if !strings.HasSuffix(pattern, ".service") {
+			if !strings.HasSuffix(pattern, "timer") && !strings.HasSuffix(pattern, ".service") {
 				pattern += ".service"
 			}
 			patterns = append(patterns, pattern)

agent/systemd_test.go

@@ -156,20 +156,23 @@ func TestGetServicePatterns(t *testing.T) {
 			expected:       []string{"*nginx*.service", "*apache*.service"},
 			cleanupEnvVars: true,
 		},
+		{
+			name:           "opt into timer monitoring",
+			prefixedEnv:    "nginx.service,docker,apache.timer",
+			unprefixedEnv:  "",
+			expected:       []string{"nginx.service", "docker.service", "apache.timer"},
+			cleanupEnvVars: true,
+		},
 	}
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
-			// Clean up any existing env vars
-			os.Unsetenv("BESZEL_AGENT_SERVICE_PATTERNS")
-			os.Unsetenv("SERVICE_PATTERNS")
-
 			// Set up environment variables
 			if tt.prefixedEnv != "" {
-				os.Setenv("BESZEL_AGENT_SERVICE_PATTERNS", tt.prefixedEnv)
+				t.Setenv("BESZEL_AGENT_SERVICE_PATTERNS", tt.prefixedEnv)
 			}
 			if tt.unprefixedEnv != "" {
-				os.Setenv("SERVICE_PATTERNS", tt.unprefixedEnv)
+				t.Setenv("SERVICE_PATTERNS", tt.unprefixedEnv)
 			}
 
 			// Run the function
@@ -177,12 +180,6 @@ func TestGetServicePatterns(t *testing.T) {
 
 			// Verify results
 			assert.Equal(t, tt.expected, result, "Patterns should match expected values")
-
-			// Cleanup
-			if tt.cleanupEnvVars {
-				os.Unsetenv("BESZEL_AGENT_SERVICE_PATTERNS")
-				os.Unsetenv("SERVICE_PATTERNS")
-			}
 		})
 	}
 }

agent/test-data/amdgpu.ids

@@ -0,0 +1,700 @@
+# List of AMDGPU IDs
+#
+# Syntax:
+# device_id,	revision_id,	product_name        <-- single tab after comma
+
+1.0.0
+1114,	C2,	AMD Radeon 860M Graphics
+1114,	C3,	AMD Radeon 840M Graphics
+1114,	D2,	AMD Radeon 860M Graphics
+1114,	D3,	AMD Radeon 840M Graphics
+1309,	00,	AMD Radeon R7 Graphics
+130A,	00,	AMD Radeon R6 Graphics
+130B,	00,	AMD Radeon R4 Graphics
+130C,	00,	AMD Radeon R7 Graphics
+130D,	00,	AMD Radeon R6 Graphics
+130E,	00,	AMD Radeon R5 Graphics
+130F,	00,	AMD Radeon R7 Graphics
+130F,	D4,	AMD Radeon R7 Graphics
+130F,	D5,	AMD Radeon R7 Graphics
+130F,	D6,	AMD Radeon R7 Graphics
+130F,	D7,	AMD Radeon R7 Graphics
+1313,	00,	AMD Radeon R7 Graphics
+1313,	D4,	AMD Radeon R7 Graphics
+1313,	D5,	AMD Radeon R7 Graphics
+1313,	D6,	AMD Radeon R7 Graphics
+1315,	00,	AMD Radeon R5 Graphics
+1315,	D4,	AMD Radeon R5 Graphics
+1315,	D5,	AMD Radeon R5 Graphics
+1315,	D6,	AMD Radeon R5 Graphics
+1315,	D7,	AMD Radeon R5 Graphics
+1316,	00,	AMD Radeon R5 Graphics
+1318,	00,	AMD Radeon R5 Graphics
+131B,	00,	AMD Radeon R4 Graphics
+131C,	00,	AMD Radeon R7 Graphics
+131D,	00,	AMD Radeon R6 Graphics
+1435,	AE,	AMD Custom GPU 0932
+1506,	C1,	AMD Radeon 610M
+1506,	C2,	AMD Radeon 610M
+1506,	C3,	AMD Radeon 610M
+1506,	C4,	AMD Radeon 610M
+150E,	C1,	AMD Radeon 890M Graphics
+150E,	C4,	AMD Radeon 890M Graphics
+150E,	C5,	AMD Radeon 890M Graphics
+150E,	C6,	AMD Radeon 890M Graphics
+150E,	D1,	AMD Radeon 890M Graphics
+150E,	D2,	AMD Radeon 890M Graphics
+150E,	D3,	AMD Radeon 890M Graphics
+1586,	C1,	Radeon 8060S Graphics
+1586,	C2,	Radeon 8050S Graphics
+1586,	C4,	Radeon 8050S Graphics
+1586,	D1,	Radeon 8060S Graphics
+1586,	D2,	Radeon 8050S Graphics
+1586,	D4,	Radeon 8050S Graphics
+1586,	D5,	Radeon 8040S Graphics
+15BF,	00,	AMD Radeon 780M Graphics
+15BF,	01,	AMD Radeon 760M Graphics
+15BF,	02,	AMD Radeon 780M Graphics
+15BF,	03,	AMD Radeon 760M Graphics
+15BF,	C1,	AMD Radeon 780M Graphics
+15BF,	C2,	AMD Radeon 780M Graphics
+15BF,	C3,	AMD Radeon 760M Graphics
+15BF,	C4,	AMD Radeon 780M Graphics
+15BF,	C5,	AMD Radeon 740M Graphics
+15BF,	C6,	AMD Radeon 780M Graphics
+15BF,	C7,	AMD Radeon 780M Graphics
+15BF,	C8,	AMD Radeon 760M Graphics
+15BF,	C9,	AMD Radeon 780M Graphics
+15BF,	CA,	AMD Radeon 740M Graphics
+15BF,	CB,	AMD Radeon 760M Graphics
+15BF,	CC,	AMD Radeon 740M Graphics
+15BF,	CD,	AMD Radeon 760M Graphics
+15BF,	CF,	AMD Radeon 780M Graphics
+15BF,	D0,	AMD Radeon 780M Graphics
+15BF,	D1,	AMD Radeon 780M Graphics
+15BF,	D2,	AMD Radeon 780M Graphics
+15BF,	D3,	AMD Radeon 780M Graphics
+15BF,	D4,	AMD Radeon 780M Graphics
+15BF,	D5,	AMD Radeon 760M Graphics
+15BF,	D6,	AMD Radeon 760M Graphics
+15BF,	D7,	AMD Radeon 780M Graphics
+15BF,	D8,	AMD Radeon 740M Graphics
+15BF,	D9,	AMD Radeon 780M Graphics
+15BF,	DA,	AMD Radeon 780M Graphics
+15BF,	DB,	AMD Radeon 760M Graphics
+15BF,	DC,	AMD Radeon 760M Graphics
+15BF,	DD,	AMD Radeon 780M Graphics
+15BF,	DE,	AMD Radeon 740M Graphics
+15BF,	DF,	AMD Radeon 760M Graphics
+15BF,	F0,	AMD Radeon 760M Graphics
+15C8,	C1,	AMD Radeon 740M Graphics
+15C8,	C2,	AMD Radeon 740M Graphics
+15C8,	C3,	AMD Radeon 740M Graphics
+15C8,	C4,	AMD Radeon 740M Graphics
+15C8,	D1,	AMD Radeon 740M Graphics
+15C8,	D2,	AMD Radeon 740M Graphics
+15C8,	D3,	AMD Radeon 740M Graphics
+15C8,	D4,	AMD Radeon 740M Graphics
+15D8,	00,	AMD Radeon RX Vega 8 Graphics WS
+15D8,	91,	AMD Radeon Vega 3 Graphics
+15D8,	91,	AMD Ryzen Embedded R1606G with Radeon Vega Gfx
+15D8,	92,	AMD Radeon Vega 3 Graphics
+15D8,	92,	AMD Ryzen Embedded R1505G with Radeon Vega Gfx
+15D8,	93,	AMD Radeon Vega 1 Graphics
+15D8,	A1,	AMD Radeon Vega 10 Graphics
+15D8,	A2,	AMD Radeon Vega 8 Graphics
+15D8,	A3,	AMD Radeon Vega 6 Graphics
+15D8,	A4,	AMD Radeon Vega 3 Graphics
+15D8,	B1,	AMD Radeon Vega 10 Graphics
+15D8,	B2,	AMD Radeon Vega 8 Graphics
+15D8,	B3,	AMD Radeon Vega 6 Graphics
+15D8,	B4,	AMD Radeon Vega 3 Graphics
+15D8,	C1,	AMD Radeon Vega 10 Graphics
+15D8,	C2,	AMD Radeon Vega 8 Graphics
+15D8,	C3,	AMD Radeon Vega 6 Graphics
+15D8,	C4,	AMD Radeon Vega 3 Graphics
+15D8,	C5,	AMD Radeon Vega 3 Graphics
+15D8,	C8,	AMD Radeon Vega 11 Graphics
+15D8,	C9,	AMD Radeon Vega 8 Graphics
+15D8,	CA,	AMD Radeon Vega 11 Graphics
+15D8,	CB,	AMD Radeon Vega 8 Graphics
+15D8,	CC,	AMD Radeon Vega 3 Graphics
+15D8,	CE,	AMD Radeon Vega 3 Graphics
+15D8,	CF,	AMD Ryzen Embedded R1305G with Radeon Vega Gfx
+15D8,	D1,	AMD Radeon Vega 10 Graphics
+15D8,	D2,	AMD Radeon Vega 8 Graphics
+15D8,	D3,	AMD Radeon Vega 6 Graphics
+15D8,	D4,	AMD Radeon Vega 3 Graphics
+15D8,	D8,	AMD Radeon Vega 11 Graphics
+15D8,	D9,	AMD Radeon Vega 8 Graphics
+15D8,	DA,	AMD Radeon Vega 11 Graphics
+15D8,	DB,	AMD Radeon Vega 3 Graphics
+15D8,	DB,	AMD Radeon Vega 8 Graphics
+15D8,	DC,	AMD Radeon Vega 3 Graphics
+15D8,	DD,	AMD Radeon Vega 3 Graphics
+15D8,	DE,	AMD Radeon Vega 3 Graphics
+15D8,	DF,	AMD Radeon Vega 3 Graphics
+15D8,	E3,	AMD Radeon Vega 3 Graphics
+15D8,	E4,	AMD Ryzen Embedded R1102G with Radeon Vega Gfx
+15DD,	81,	AMD Ryzen Embedded V1807B with Radeon Vega Gfx
+15DD,	82,	AMD Ryzen Embedded V1756B with Radeon Vega Gfx
+15DD,	83,	AMD Ryzen Embedded V1605B with Radeon Vega Gfx
+15DD,	84,	AMD Radeon Vega 6 Graphics
+15DD,	85,	AMD Ryzen Embedded V1202B with Radeon Vega Gfx
+15DD,	86,	AMD Radeon Vega 11 Graphics
+15DD,	88,	AMD Radeon Vega 8 Graphics
+15DD,	C1,	AMD Radeon Vega 11 Graphics
+15DD,	C2,	AMD Radeon Vega 8 Graphics
+15DD,	C3,	AMD Radeon Vega 3 / 10 Graphics
+15DD,	C4,	AMD Radeon Vega 8 Graphics
+15DD,	C5,	AMD Radeon Vega 3 Graphics
+15DD,	C6,	AMD Radeon Vega 11 Graphics
+15DD,	C8,	AMD Radeon Vega 8 Graphics
+15DD,	C9,	AMD Radeon Vega 11 Graphics
+15DD,	CA,	AMD Radeon Vega 8 Graphics
+15DD,	CB,	AMD Radeon Vega 3 Graphics
+15DD,	CC,	AMD Radeon Vega 6 Graphics
+15DD,	CE,	AMD Radeon Vega 3 Graphics
+15DD,	CF,	AMD Radeon Vega 3 Graphics
+15DD,	D0,	AMD Radeon Vega 10 Graphics
+15DD,	D1,	AMD Radeon Vega 8 Graphics
+15DD,	D3,	AMD Radeon Vega 11 Graphics
+15DD,	D5,	AMD Radeon Vega 8 Graphics
+15DD,	D6,	AMD Radeon Vega 11 Graphics
+15DD,	D7,	AMD Radeon Vega 8 Graphics
+15DD,	D8,	AMD Radeon Vega 3 Graphics
+15DD,	D9,	AMD Radeon Vega 6 Graphics
+15DD,	E1,	AMD Radeon Vega 3 Graphics
+15DD,	E2,	AMD Radeon Vega 3 Graphics
+163F,	AE,	AMD Custom GPU 0405
+163F,	E1,	AMD Custom GPU 0405
+164E,	D8,	AMD Radeon 610M
+164E,	D9,	AMD Radeon 610M
+164E,	DA,	AMD Radeon 610M
+164E,	DB,	AMD Radeon 610M
+164E,	DC,	AMD Radeon 610M
+1681,	06,	AMD Radeon 680M
+1681,	07,	AMD Radeon 660M
+1681,	0A,	AMD Radeon 680M
+1681,	0B,	AMD Radeon 660M
+1681,	C7,	AMD Radeon 680M
+1681,	C8,	AMD Radeon 680M
+1681,	C9,	AMD Radeon 660M
+1900,	01,	AMD Radeon 780M Graphics
+1900,	02,	AMD Radeon 760M Graphics
+1900,	03,	AMD Radeon 780M Graphics
+1900,	04,	AMD Radeon 760M Graphics
+1900,	05,	AMD Radeon 780M Graphics
+1900,	06,	AMD Radeon 780M Graphics
+1900,	07,	AMD Radeon 760M Graphics
+1900,	B0,	AMD Radeon 780M Graphics
+1900,	B1,	AMD Radeon 780M Graphics
+1900,	B2,	AMD Radeon 780M Graphics
+1900,	B3,	AMD Radeon 780M Graphics
+1900,	B4,	AMD Radeon 780M Graphics
+1900,	B5,	AMD Radeon 780M Graphics
+1900,	B6,	AMD Radeon 780M Graphics
+1900,	B7,	AMD Radeon 760M Graphics
+1900,	B8,	AMD Radeon 760M Graphics
+1900,	B9,	AMD Radeon 780M Graphics
+1900,	BA,	AMD Radeon 780M Graphics
+1900,	BB,	AMD Radeon 780M Graphics
+1900,	C0,	AMD Radeon 780M Graphics
+1900,	C1,	AMD Radeon 760M Graphics
+1900,	C2,	AMD Radeon 780M Graphics
+1900,	C3,	AMD Radeon 760M Graphics
+1900,	C4,	AMD Radeon 780M Graphics
+1900,	C5,	AMD Radeon 780M Graphics
+1900,	C6,	AMD Radeon 760M Graphics
+1900,	C7,	AMD Radeon 780M Graphics
+1900,	C8,	AMD Radeon 760M Graphics
+1900,	C9,	AMD Radeon 780M Graphics
+1900,	CA,	AMD Radeon 760M Graphics
+1900,	CB,	AMD Radeon 780M Graphics
+1900,	CC,	AMD Radeon 780M Graphics
+1900,	CD,	AMD Radeon 760M Graphics
+1900,	CE,	AMD Radeon 780M Graphics
+1900,	CF,	AMD Radeon 760M Graphics
+1900,	D0,	AMD Radeon 780M Graphics
+1900,	D1,	AMD Radeon 760M Graphics
+1900,	D2,	AMD Radeon 780M Graphics
+1900,	D3,	AMD Radeon 760M Graphics
+1900,	D4,	AMD Radeon 780M Graphics
+1900,	D5,	AMD Radeon 780M Graphics
+1900,	D6,	AMD Radeon 760M Graphics
+1900,	D7,	AMD Radeon 780M Graphics
+1900,	D8,	AMD Radeon 760M Graphics
+1900,	D9,	AMD Radeon 780M Graphics
+1900,	DA,	AMD Radeon 760M Graphics
+1900,	DB,	AMD Radeon 780M Graphics
+1900,	DC,	AMD Radeon 780M Graphics
+1900,	DD,	AMD Radeon 760M Graphics
+1900,	DE,	AMD Radeon 780M Graphics
+1900,	DF,	AMD Radeon 760M Graphics
+1900,	F0,	AMD Radeon 780M Graphics
+1900,	F1,	AMD Radeon 780M Graphics
+1900,	F2,	AMD Radeon 780M Graphics
+1901,	C1,	AMD Radeon 740M Graphics
+1901,	C2,	AMD Radeon 740M Graphics
+1901,	C3,	AMD Radeon 740M Graphics
+1901,	C6,	AMD Radeon 740M Graphics
+1901,	C7,	AMD Radeon 740M Graphics
+1901,	C8,	AMD Radeon 740M Graphics
+1901,	C9,	AMD Radeon 740M Graphics
+1901,	CA,	AMD Radeon 740M Graphics
+1901,	D1,	AMD Radeon 740M Graphics
+1901,	D2,	AMD Radeon 740M Graphics
+1901,	D3,	AMD Radeon 740M Graphics
+1901,	D4,	AMD Radeon 740M Graphics
+1901,	D5,	AMD Radeon 740M Graphics
+1901,	D6,	AMD Radeon 740M Graphics
+1901,	D7,	AMD Radeon 740M Graphics
+1901,	D8,	AMD Radeon 740M Graphics
+6600,	00,	AMD Radeon HD 8600 / 8700M
+6600,	81,	AMD Radeon R7 M370
+6601,	00,	AMD Radeon HD 8500M / 8700M
+6604,	00,	AMD Radeon R7 M265 Series
+6604,	81,	AMD Radeon R7 M350
+6605,	00,	AMD Radeon R7 M260 Series
+6605,	81,	AMD Radeon R7 M340
+6606,	00,	AMD Radeon HD 8790M
+6607,	00,	AMD Radeon R5 M240
+6608,	00,	AMD FirePro W2100
+6610,	00,	AMD Radeon R7 200 Series
+6610,	81,	AMD Radeon R7 350
+6610,	83,	AMD Radeon R5 340
+6610,	87,	AMD Radeon R7 200 Series
+6611,	00,	AMD Radeon R7 200 Series
+6611,	87,	AMD Radeon R7 200 Series
+6613,	00,	AMD Radeon R7 200 Series
+6617,	00,	AMD Radeon R7 240 Series
+6617,	87,	AMD Radeon R7 200 Series
+6617,	C7,	AMD Radeon R7 240 Series
+6640,	00,	AMD Radeon HD 8950
+6640,	80,	AMD Radeon R9 M380
+6646,	00,	AMD Radeon R9 M280X
+6646,	80,	AMD Radeon R9 M385
+6646,	80,	AMD Radeon R9 M470X
+6647,	00,	AMD Radeon R9 M200X Series
+6647,	80,	AMD Radeon R9 M380
+6649,	00,	AMD FirePro W5100
+6658,	00,	AMD Radeon R7 200 Series
+665C,	00,	AMD Radeon HD 7700 Series
+665D,	00,	AMD Radeon R7 200 Series
+665F,	81,	AMD Radeon R7 360 Series
+6660,	00,	AMD Radeon HD 8600M Series
+6660,	81,	AMD Radeon R5 M335
+6660,	83,	AMD Radeon R5 M330
+6663,	00,	AMD Radeon HD 8500M Series
+6663,	83,	AMD Radeon R5 M320
+6664,	00,	AMD Radeon R5 M200 Series
+6665,	00,	AMD Radeon R5 M230 Series
+6665,	83,	AMD Radeon R5 M320
+6665,	C3,	AMD Radeon R5 M435
+6666,	00,	AMD Radeon R5 M200 Series
+6667,	00,	AMD Radeon R5 M200 Series
+666F,	00,	AMD Radeon HD 8500M
+66A1,	02,	AMD Instinct MI60 / MI50
+66A1,	06,	AMD Radeon Pro VII
+66AF,	C1,	AMD Radeon VII
+6780,	00,	AMD FirePro W9000
+6784,	00,	ATI FirePro V (FireGL V) Graphics Adapter
+6788,	00,	ATI FirePro V (FireGL V) Graphics Adapter
+678A,	00,	AMD FirePro W8000
+6798,	00,	AMD Radeon R9 200 / HD 7900 Series
+6799,	00,	AMD Radeon HD 7900 Series
+679A,	00,	AMD Radeon HD 7900 Series
+679B,	00,	AMD Radeon HD 7900 Series
+679E,	00,	AMD Radeon HD 7800 Series
+67A0,	00,	AMD Radeon FirePro W9100
+67A1,	00,	AMD Radeon FirePro W8100
+67B0,	00,	AMD Radeon R9 200 Series
+67B0,	80,	AMD Radeon R9 390 Series
+67B1,	00,	AMD Radeon R9 200 Series
+67B1,	80,	AMD Radeon R9 390 Series
+67B9,	00,	AMD Radeon R9 200 Series
+67C0,	00,	AMD Radeon Pro WX 7100 Graphics
+67C0,	80,	AMD Radeon E9550
+67C2,	01,	AMD Radeon Pro V7350x2
+67C2,	02,	AMD Radeon Pro V7300X
+67C4,	00,	AMD Radeon Pro WX 7100 Graphics
+67C4,	80,	AMD Radeon E9560 / E9565 Graphics
+67C7,	00,	AMD Radeon Pro WX 5100 Graphics
+67C7,	80,	AMD Radeon E9390 Graphics
+67D0,	01,	AMD Radeon Pro V7350x2
+67D0,	02,	AMD Radeon Pro V7300X
+67DF,	C0,	AMD Radeon Pro 580X
+67DF,	C1,	AMD Radeon RX 580 Series
+67DF,	C2,	AMD Radeon RX 570 Series
+67DF,	C3,	AMD Radeon RX 580 Series
+67DF,	C4,	AMD Radeon RX 480 Graphics
+67DF,	C5,	AMD Radeon RX 470 Graphics
+67DF,	C6,	AMD Radeon RX 570 Series
+67DF,	C7,	AMD Radeon RX 480 Graphics
+67DF,	CF,	AMD Radeon RX 470 Graphics
+67DF,	D7,	AMD Radeon RX 470 Graphics
+67DF,	E0,	AMD Radeon RX 470 Series
+67DF,	E1,	AMD Radeon RX 590 Series
+67DF,	E3,	AMD Radeon RX Series
+67DF,	E7,	AMD Radeon RX 580 Series
+67DF,	EB,	AMD Radeon Pro 580X
+67DF,	EF,	AMD Radeon RX 570 Series
+67DF,	F7,	AMD Radeon RX P30PH
+67DF,	FF,	AMD Radeon RX 470 Series
+67E0,	00,	AMD Radeon Pro WX Series
+67E3,	00,	AMD Radeon Pro WX 4100
+67E8,	00,	AMD Radeon Pro WX Series
+67E8,	01,	AMD Radeon Pro WX Series
+67E8,	80,	AMD Radeon E9260 Graphics
+67EB,	00,	AMD Radeon Pro V5300X
+67EF,	C0,	AMD Radeon RX Graphics
+67EF,	C1,	AMD Radeon RX 460 Graphics
+67EF,	C2,	AMD Radeon Pro Series
+67EF,	C3,	AMD Radeon RX Series
+67EF,	C5,	AMD Radeon RX 460 Graphics
+67EF,	C7,	AMD Radeon RX Graphics
+67EF,	CF,	AMD Radeon RX 460 Graphics
+67EF,	E0,	AMD Radeon RX 560 Series
+67EF,	E1,	AMD Radeon RX Series
+67EF,	E2,	AMD Radeon RX 560X
+67EF,	E3,	AMD Radeon RX Series
+67EF,	E5,	AMD Radeon RX 560 Series
+67EF,	E7,	AMD Radeon RX 560 Series
+67EF,	EF,	AMD Radeon 550 Series
+67EF,	FF,	AMD Radeon RX 460 Graphics
+67FF,	C0,	AMD Radeon Pro 465
+67FF,	C1,	AMD Radeon RX 560 Series
+67FF,	CF,	AMD Radeon RX 560 Series
+67FF,	EF,	AMD Radeon RX 560 Series
+67FF,	FF,	AMD Radeon RX 550 Series
+6800,	00,	AMD Radeon HD 7970M
+6801,	00,	AMD Radeon HD 8970M
+6806,	00,	AMD Radeon R9 M290X
+6808,	00,	AMD FirePro W7000
+6808,	00,	ATI FirePro V (FireGL V) Graphics Adapter
+6809,	00,	ATI FirePro W5000
+6810,	00,	AMD Radeon R9 200 Series
+6810,	81,	AMD Radeon R9 370 Series
+6811,	00,	AMD Radeon R9 200 Series
+6811,	81,	AMD Radeon R7 370 Series
+6818,	00,	AMD Radeon HD 7800 Series
+6819,	00,	AMD Radeon HD 7800 Series
+6820,	00,	AMD Radeon R9 M275X
+6820,	81,	AMD Radeon R9 M375
+6820,	83,	AMD Radeon R9 M375X
+6821,	00,	AMD Radeon R9 M200X Series
+6821,	83,	AMD Radeon R9 M370X
+6821,	87,	AMD Radeon R7 M380
+6822,	00,	AMD Radeon E8860
+6823,	00,	AMD Radeon R9 M200X Series
+6825,	00,	AMD Radeon HD 7800M Series
+6826,	00,	AMD Radeon HD 7700M Series
+6827,	00,	AMD Radeon HD 7800M Series
+6828,	00,	AMD FirePro W600
+682B,	00,	AMD Radeon HD 8800M Series
+682B,	87,	AMD Radeon R9 M360
+682C,	00,	AMD FirePro W4100
+682D,	00,	AMD Radeon HD 7700M Series
+682F,	00,	AMD Radeon HD 7700M Series
+6830,	00,	AMD Radeon 7800M Series
+6831,	00,	AMD Radeon 7700M Series
+6835,	00,	AMD Radeon R7 Series / HD 9000 Series
+6837,	00,	AMD Radeon HD 7700 Series
+683D,	00,	AMD Radeon HD 7700 Series
+683F,	00,	AMD Radeon HD 7700 Series
+684C,	00,	ATI FirePro V (FireGL V) Graphics Adapter
+6860,	00,	AMD Radeon Instinct MI25
+6860,	01,	AMD Radeon Instinct MI25
+6860,	02,	AMD Radeon Instinct MI25
+6860,	03,	AMD Radeon Pro V340
+6860,	04,	AMD Radeon Instinct MI25x2
+6860,	07,	AMD Radeon Pro V320
+6861,	00,	AMD Radeon Pro WX 9100
+6862,	00,	AMD Radeon Pro SSG
+6863,	00,	AMD Radeon Vega Frontier Edition
+6864,	03,	AMD Radeon Pro V340
+6864,	04,	AMD Radeon Instinct MI25x2
+6864,	05,	AMD Radeon Pro V340
+6868,	00,	AMD Radeon Pro WX 8200
+686C,	00,	AMD Radeon Instinct MI25 MxGPU
+686C,	01,	AMD Radeon Instinct MI25 MxGPU
+686C,	02,	AMD Radeon Instinct MI25 MxGPU
+686C,	03,	AMD Radeon Pro V340 MxGPU
+686C,	04,	AMD Radeon Instinct MI25x2 MxGPU
+686C,	05,	AMD Radeon Pro V340L MxGPU
+686C,	06,	AMD Radeon Instinct MI25 MxGPU
+687F,	01,	AMD Radeon RX Vega
+687F,	C0,	AMD Radeon RX Vega
+687F,	C1,	AMD Radeon RX Vega
+687F,	C3,	AMD Radeon RX Vega
+687F,	C7,	AMD Radeon RX Vega
+6900,	00,	AMD Radeon R7 M260
+6900,	81,	AMD Radeon R7 M360
+6900,	83,	AMD Radeon R7 M340
+6900,	C1,	AMD Radeon R5 M465 Series
+6900,	C3,	AMD Radeon R5 M445 Series
+6900,	D1,	AMD Radeon 530 Series
+6900,	D3,	AMD Radeon 530 Series
+6901,	00,	AMD Radeon R5 M255
+6902,	00,	AMD Radeon Series
+6907,	00,	AMD Radeon R5 M255
+6907,	87,	AMD Radeon R5 M315
+6920,	00,	AMD Radeon R9 M395X
+6920,	01,	AMD Radeon R9 M390X
+6921,	00,	AMD Radeon R9 M390X
+6929,	00,	AMD FirePro S7150
+6929,	01,	AMD FirePro S7100X
+692B,	00,	AMD FirePro W7100
+6938,	00,	AMD Radeon R9 200 Series
+6938,	F0,	AMD Radeon R9 200 Series
+6938,	F1,	AMD Radeon R9 380 Series
+6939,	00,	AMD Radeon R9 200 Series
+6939,	F0,	AMD Radeon R9 200 Series
+6939,	F1,	AMD Radeon R9 380 Series
+694C,	C0,	AMD Radeon RX Vega M GH Graphics
+694E,	C0,	AMD Radeon RX Vega M GL Graphics
+6980,	00,	AMD Radeon Pro WX 3100
+6981,	00,	AMD Radeon Pro WX 3200 Series
+6981,	01,	AMD Radeon Pro WX 3200 Series
+6981,	10,	AMD Radeon Pro WX 3200 Series
+6985,	00,	AMD Radeon Pro WX 3100
+6986,	00,	AMD Radeon Pro WX 2100
+6987,	80,	AMD Embedded Radeon E9171
+6987,	C0,	AMD Radeon 550X Series
+6987,	C1,	AMD Radeon RX 640
+6987,	C3,	AMD Radeon 540X Series
+6987,	C7,	AMD Radeon 540
+6995,	00,	AMD Radeon Pro WX 2100
+6997,	00,	AMD Radeon Pro WX 2100
+699F,	81,	AMD Embedded Radeon E9170 Series
+699F,	C0,	AMD Radeon 500 Series
+699F,	C1,	AMD Radeon 540 Series
+699F,	C3,	AMD Radeon 500 Series
+699F,	C7,	AMD Radeon RX 550 / 550 Series
+699F,	C9,	AMD Radeon 540
+6FDF,	E7,	AMD Radeon RX 590 GME
+6FDF,	EF,	AMD Radeon RX 580 2048SP
+7300,	C1,	AMD FirePro S9300 x2
+7300,	C8,	AMD Radeon R9 Fury Series
+7300,	C9,	AMD Radeon Pro Duo
+7300,	CA,	AMD Radeon R9 Fury Series
+7300,	CB,	AMD Radeon R9 Fury Series
+7312,	00,	AMD Radeon Pro W5700
+731E,	C6,	AMD Radeon RX 5700XTB
+731E,	C7,	AMD Radeon RX 5700B
+731F,	C0,	AMD Radeon RX 5700 XT 50th Anniversary
+731F,	C1,	AMD Radeon RX 5700 XT
+731F,	C2,	AMD Radeon RX 5600M
+731F,	C3,	AMD Radeon RX 5700M
+731F,	C4,	AMD Radeon RX 5700
+731F,	C5,	AMD Radeon RX 5700 XT
+731F,	CA,	AMD Radeon RX 5600 XT
+731F,	CB,	AMD Radeon RX 5600 OEM
+7340,	C1,	AMD Radeon RX 5500M
+7340,	C3,	AMD Radeon RX 5300M
+7340,	C5,	AMD Radeon RX 5500 XT
+7340,	C7,	AMD Radeon RX 5500
+7340,	C9,	AMD Radeon RX 5500XTB
+7340,	CF,	AMD Radeon RX 5300
+7341,	00,	AMD Radeon Pro W5500
+7347,	00,	AMD Radeon Pro W5500M
+7360,	41,	AMD Radeon Pro 5600M
+7360,	C3,	AMD Radeon Pro V520
+7362,	C1,	AMD Radeon Pro V540
+7362,	C3,	AMD Radeon Pro V520
+738C,	01,	AMD Instinct MI100
+73A1,	00,	AMD Radeon Pro V620
+73A3,	00,	AMD Radeon Pro W6800
+73A5,	C0,	AMD Radeon RX 6950 XT
+73AE,	00,	AMD Radeon Pro V620 MxGPU
+73AF,	C0,	AMD Radeon RX 6900 XT
+73BF,	C0,	AMD Radeon RX 6900 XT
+73BF,	C1,	AMD Radeon RX 6800 XT
+73BF,	C3,	AMD Radeon RX 6800
+73DF,	C0,	AMD Radeon RX 6750 XT
+73DF,	C1,	AMD Radeon RX 6700 XT
+73DF,	C2,	AMD Radeon RX 6800M
+73DF,	C3,	AMD Radeon RX 6800M
+73DF,	C5,	AMD Radeon RX 6700 XT
+73DF,	CF,	AMD Radeon RX 6700M
+73DF,	D5,	AMD Radeon RX 6750 GRE 12GB
+73DF,	D7,	AMD TDC-235
+73DF,	DF,	AMD Radeon RX 6700
+73DF,	E5,	AMD Radeon RX 6750 GRE 12GB
+73DF,	FF,	AMD Radeon RX 6700
+73E0,	00,	AMD Radeon RX 6600M
+73E1,	00,	AMD Radeon Pro W6600M
+73E3,	00,	AMD Radeon Pro W6600
+73EF,	C0,	AMD Radeon RX 6800S
+73EF,	C1,	AMD Radeon RX 6650 XT
+73EF,	C2,	AMD Radeon RX 6700S
+73EF,	C3,	AMD Radeon RX 6650M
+73EF,	C4,	AMD Radeon RX 6650M XT
+73FF,	C1,	AMD Radeon RX 6600 XT
+73FF,	C3,	AMD Radeon RX 6600M
+73FF,	C7,	AMD Radeon RX 6600
+73FF,	CB,	AMD Radeon RX 6600S
+73FF,	CF,	AMD Radeon RX 6600 LE
+73FF,	DF,	AMD Radeon RX 6750 GRE 10GB
+7408,	00,	AMD Instinct MI250X
+740C,	01,	AMD Instinct MI250X / MI250
+740F,	02,	AMD Instinct MI210
+7421,	00,	AMD Radeon Pro W6500M
+7422,	00,	AMD Radeon Pro W6400
+7423,	00,	AMD Radeon Pro W6300M
+7423,	01,	AMD Radeon Pro W6300
+7424,	00,	AMD Radeon RX 6300
+743F,	C1,	AMD Radeon RX 6500 XT
+743F,	C3,	AMD Radeon RX 6500
+743F,	C3,	AMD Radeon RX 6500M
+743F,	C7,	AMD Radeon RX 6400
+743F,	C8,	AMD Radeon RX 6500M
+743F,	CC,	AMD Radeon 6550S
+743F,	CE,	AMD Radeon RX 6450M
+743F,	CF,	AMD Radeon RX 6300M
+743F,	D3,	AMD Radeon RX 6550M
+743F,	D7,	AMD Radeon RX 6400
+7448,	00,	AMD Radeon Pro W7900
+7449,	00,	AMD Radeon Pro W7800 48GB
+744A,	00,	AMD Radeon Pro W7900 Dual Slot
+744B,	00,	AMD Radeon Pro W7900D
+744C,	C8,	AMD Radeon RX 7900 XTX
+744C,	CC,	AMD Radeon RX 7900 XT
+744C,	CE,	AMD Radeon RX 7900 GRE
+744C,	CF,	AMD Radeon RX 7900M
+745E,	CC,	AMD Radeon Pro W7800
+7460,	00,	AMD Radeon Pro V710
+7461,	00,	AMD Radeon Pro V710 MxGPU
+7470,	00,	AMD Radeon Pro W7700
+747E,	C8,	AMD Radeon RX 7800 XT
+747E,	D8,	AMD Radeon RX 7800M
+747E,	DB,	AMD Radeon RX 7700
+747E,	FF,	AMD Radeon RX 7700 XT
+7480,	00,	AMD Radeon Pro W7600
+7480,	C0,	AMD Radeon RX 7600 XT
+7480,	C1,	AMD Radeon RX 7700S
+7480,	C2,	AMD Radeon RX 7650 GRE
+7480,	C3,	AMD Radeon RX 7600S
+7480,	C7,	AMD Radeon RX 7600M XT
+7480,	CF,	AMD Radeon RX 7600
+7481,   C7,     AMD Steam Machine
+7483,	CF,	AMD Radeon RX 7600M
+7489,	00,	AMD Radeon Pro W7500
+7499,	00,	AMD Radeon Pro W7400
+7499,	C0,	AMD Radeon RX 7400
+7499,	C1,	AMD Radeon RX 7300
+74A0,	00,	AMD Instinct MI300A
+74A1,	00,	AMD Instinct MI300X
+74A2,	00,	AMD Instinct MI308X
+74A5,	00,	AMD Instinct MI325X
+74A8,	00,	AMD Instinct MI308X HF
+74A9,	00,	AMD Instinct MI300X HF
+74B5,	00,	AMD Instinct MI300X VF
+74B6,	00,	AMD Instinct MI308X
+74BD,	00,	AMD Instinct MI300X HF
+7550,	C0,	AMD Radeon RX 9070 XT
+7550,	C2,	AMD Radeon RX 9070 GRE
+7550,	C3,	AMD Radeon RX 9070
+7551,	C0,	AMD Radeon AI PRO R9700
+7590,	C0,	AMD Radeon RX 9060 XT
+7590,	C7,	AMD Radeon RX 9060
+75A0,	C0,	AMD Instinct MI350X
+75A3,	C0,	AMD Instinct MI355X
+75B0,	C0,	AMD Instinct MI350X VF
+75B3,	C0,	AMD Instinct MI355X VF
+9830,	00,	AMD Radeon HD 8400 / R3 Series
+9831,	00,	AMD Radeon HD 8400E
+9832,	00,	AMD Radeon HD 8330
+9833,	00,	AMD Radeon HD 8330E
+9834,	00,	AMD Radeon HD 8210
+9835,	00,	AMD Radeon HD 8210E
+9836,	00,	AMD Radeon HD 8200 / R3 Series
+9837,	00,	AMD Radeon HD 8280E
+9838,	00,	AMD Radeon HD 8200 / R3 series
+9839,	00,	AMD Radeon HD 8180
+983D,	00,	AMD Radeon HD 8250
+9850,	00,	AMD Radeon R3 Graphics
+9850,	03,	AMD Radeon R3 Graphics
+9850,	40,	AMD Radeon R2 Graphics
+9850,	45,	AMD Radeon R3 Graphics
+9851,	00,	AMD Radeon R4 Graphics
+9851,	01,	AMD Radeon R5E Graphics
+9851,	05,	AMD Radeon R5 Graphics
+9851,	06,	AMD Radeon R5E Graphics
+9851,	40,	AMD Radeon R4 Graphics
+9851,	45,	AMD Radeon R5 Graphics
+9852,	00,	AMD Radeon R2 Graphics
+9852,	40,	AMD Radeon E1 Graphics
+9853,	00,	AMD Radeon R2 Graphics
+9853,	01,	AMD Radeon R4E Graphics
+9853,	03,	AMD Radeon R2 Graphics
+9853,	05,	AMD Radeon R1E Graphics
+9853,	06,	AMD Radeon R1E Graphics
+9853,	07,	AMD Radeon R1E Graphics
+9853,	08,	AMD Radeon R1E Graphics
+9853,	40,	AMD Radeon R2 Graphics
+9854,	00,	AMD Radeon R3 Graphics
+9854,	01,	AMD Radeon R3E Graphics
+9854,	02,	AMD Radeon R3 Graphics
+9854,	05,	AMD Radeon R2 Graphics
+9854,	06,	AMD Radeon R4 Graphics
+9854,	07,	AMD Radeon R3 Graphics
+9855,	02,	AMD Radeon R6 Graphics
+9855,	05,	AMD Radeon R4 Graphics
+9856,	00,	AMD Radeon R2 Graphics
+9856,	01,	AMD Radeon R2E Graphics
+9856,	02,	AMD Radeon R2 Graphics
+9856,	05,	AMD Radeon R1E Graphics
+9856,	06,	AMD Radeon R2 Graphics
+9856,	07,	AMD Radeon R1E Graphics
+9856,	08,	AMD Radeon R1E Graphics
+9856,	13,	AMD Radeon R1E Graphics
+9874,	81,	AMD Radeon R6 Graphics
+9874,	84,	AMD Radeon R7 Graphics
+9874,	85,	AMD Radeon R6 Graphics
+9874,	87,	AMD Radeon R5 Graphics
+9874,	88,	AMD Radeon R7E Graphics
+9874,	89,	AMD Radeon R6E Graphics
+9874,	C4,	AMD Radeon R7 Graphics
+9874,	C5,	AMD Radeon R6 Graphics
+9874,	C6,	AMD Radeon R6 Graphics
+9874,	C7,	AMD Radeon R5 Graphics
+9874,	C8,	AMD Radeon R7 Graphics
+9874,	C9,	AMD Radeon R7 Graphics
+9874,	CA,	AMD Radeon R5 Graphics
+9874,	CB,	AMD Radeon R5 Graphics
+9874,	CC,	AMD Radeon R7 Graphics
+9874,	CD,	AMD Radeon R7 Graphics
+9874,	CE,	AMD Radeon R5 Graphics
+9874,	E1,	AMD Radeon R7 Graphics
+9874,	E2,	AMD Radeon R7 Graphics
+9874,	E3,	AMD Radeon R7 Graphics
+9874,	E4,	AMD Radeon R7 Graphics
+9874,	E5,	AMD Radeon R5 Graphics
+9874,	E6,	AMD Radeon R5 Graphics
+98E4,	80,	AMD Radeon R5E Graphics
+98E4,	81,	AMD Radeon R4E Graphics
+98E4,	83,	AMD Radeon R2E Graphics
+98E4,	84,	AMD Radeon R2E Graphics
+98E4,	86,	AMD Radeon R1E Graphics
+98E4,	C0,	AMD Radeon R4 Graphics
+98E4,	C1,	AMD Radeon R5 Graphics
+98E4,	C2,	AMD Radeon R4 Graphics
+98E4,	C4,	AMD Radeon R5 Graphics
+98E4,	C6,	AMD Radeon R5 Graphics
+98E4,	C8,	AMD Radeon R4 Graphics
+98E4,	C9,	AMD Radeon R4 Graphics
+98E4,	CA,	AMD Radeon R5 Graphics
+98E4,	D0,	AMD Radeon R2 Graphics
+98E4,	D1,	AMD Radeon R2 Graphics
+98E4,	D2,	AMD Radeon R2 Graphics
+98E4,	D4,	AMD Radeon R2 Graphics
+98E4,	D9,	AMD Radeon R5 Graphics
+98E4,	DA,	AMD Radeon R5 Graphics
+98E4,	DB,	AMD Radeon R3 Graphics
+98E4,	E1,	AMD Radeon R3 Graphics
+98E4,	E2,	AMD Radeon R3 Graphics
+98E4,	E9,	AMD Radeon R4 Graphics
+98E4,	EA,	AMD Radeon R4 Graphics
+98E4,	EB,	AMD Radeon R3 Graphics
+98E4,	EB,	AMD Radeon R4 Graphics

agent/test-data/nvtop.json

@@ -0,0 +1,34 @@
+[
+  {
+   "device_name": "NVIDIA GeForce RTX 3050 Ti Laptop GPU",
+   "gpu_clock": "1485MHz",
+   "mem_clock": "6001MHz",
+   "temp": "48C",
+   "fan_speed": null,
+   "power_draw": "13W",
+   "gpu_util": "5%",
+   "encode": "0%",
+   "decode": "0%",
+   "mem_util": "8%",
+   "mem_total": "4294967296",
+   "mem_used": "349372416",
+   "mem_free": "3945594880",
+   "processes" : []
+  },
+  {
+   "device_name": "AMD Radeon 680M",
+   "gpu_clock": "2200MHz",
+   "mem_clock": "2400MHz",
+   "temp": "48C",
+   "fan_speed": "CPU Fan",
+   "power_draw": "9W",
+   "gpu_util": "12%",
+   "encode": null,
+   "decode": "0%",
+   "mem_util": "7%",
+   "mem_total": "16929173504",
+   "mem_used": "1213784064",
+   "mem_free": "15715389440",
+   "processes" : []
+  }
+]

agent/test-data/smart/apple_nvme.json

@@ -0,0 +1,51 @@
+{
+  "json_format_version": [1, 0],
+  "smartctl": {
+    "version": [7, 4],
+    "argv": ["smartctl", "-aix", "-j", "IOService:/AppleARMPE/arm-io@10F00000/AppleT810xIO/ans@77400000/AppleASCWrapV4/iop-ans-nub/RTBuddy(ANS2)/RTBuddyService/AppleANS3NVMeController/NS_01@1"],
+    "exit_status": 4
+  },
+  "device": {
+    "name": "IOService:/AppleARMPE/arm-io@10F00000/AppleT810xIO/ans@77400000/AppleASCWrapV4/iop-ans-nub/RTBuddy(ANS2)/RTBuddyService/AppleANS3NVMeController/NS_01@1",
+    "info_name": "IOService:/AppleARMPE/arm-io@10F00000/AppleT810xIO/ans@77400000/AppleASCWrapV4/iop-ans-nub/RTBuddy(ANS2)/RTBuddyService/AppleANS3NVMeController/NS_01@1",
+    "type": "nvme",
+    "protocol": "NVMe"
+  },
+  "model_name": "APPLE SSD AP0256Q",
+  "serial_number": "0ba0147940253c15",
+  "firmware_version": "555",
+  "smart_support": {
+    "available": true,
+    "enabled": true
+  },
+  "smart_status": {
+    "passed": true,
+    "nvme": {
+      "value": 0
+    }
+  },
+  "nvme_smart_health_information_log": {
+    "critical_warning": 0,
+    "temperature": 42,
+    "available_spare": 100,
+    "available_spare_threshold": 99,
+    "percentage_used": 1,
+    "data_units_read": 270189386,
+    "data_units_written": 166753862,
+    "host_reads": 7543766995,
+    "host_writes": 3761621926,
+    "controller_busy_time": 0,
+    "power_cycles": 366,
+    "power_on_hours": 2850,
+    "unsafe_shutdowns": 195,
+    "media_errors": 0,
+    "num_err_log_entries": 0
+  },
+  "temperature": {
+    "current": 42
+  },
+  "power_cycle_count": 366,
+  "power_on_time": {
+    "hours": 2850
+  }
+}

agent/update.go

@@ -63,9 +63,9 @@ func detectRestarter() restarter {
 	if path, err := exec.LookPath("rc-service"); err == nil {
 		return &openRCRestarter{cmd: path}
 	}
-    if path, err := exec.LookPath("procd"); err == nil {
-        return &openWRTRestarter{cmd: path}
-    }
+	if path, err := exec.LookPath("procd"); err == nil {
+		return &openWRTRestarter{cmd: path}
+	}
 	if path, err := exec.LookPath("service"); err == nil {
 		if runtime.GOOS == "freebsd" {
 			return &freeBSDRestarter{cmd: path}
@@ -79,7 +79,7 @@ func detectRestarter() restarter {
 func Update(useMirror bool) error {
 	exePath, _ := os.Executable()
 
-	dataDir, err := getDataDir()
+	dataDir, err := GetDataDir()
 	if err != nil {
 		dataDir = os.TempDir()
 	}
@@ -125,4 +125,3 @@ func Update(useMirror bool) error {
 
 	return nil
 }
-

agent/utils.go

@@ -1,15 +0,0 @@
-package agent
-
-import "math"
-
-func bytesToMegabytes(b float64) float64 {
-	return twoDecimals(b / 1048576)
-}
-
-func bytesToGigabytes(b uint64) float64 {
-	return twoDecimals(float64(b) / 1073741824)
-}
-
-func twoDecimals(value float64) float64 {
-	return math.Round(value*100) / 100
-}

agent/utils/utils.go

@@ -0,0 +1,117 @@
+// Package utils provides utility functions for the agent.
+package utils
+
+import (
+	"fmt"
+	"io"
+	"math"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"runtime"
+	"strconv"
+	"strings"
+)
+
+// GetEnv retrieves an environment variable with a "BESZEL_AGENT_" prefix, or falls back to the unprefixed key.
+func GetEnv(key string) (value string, exists bool) {
+	if value, exists = os.LookupEnv("BESZEL_AGENT_" + key); exists {
+		return value, exists
+	}
+	return os.LookupEnv(key)
+}
+
+// BytesToMegabytes converts bytes to megabytes and rounds to two decimal places.
+func BytesToMegabytes(b float64) float64 {
+	return TwoDecimals(b / 1048576)
+}
+
+// BytesToGigabytes converts bytes to gigabytes and rounds to two decimal places.
+func BytesToGigabytes(b uint64) float64 {
+	return TwoDecimals(float64(b) / 1073741824)
+}
+
+// TwoDecimals rounds a float64 value to two decimal places.
+func TwoDecimals(value float64) float64 {
+	return math.Round(value*100) / 100
+}
+
+// func RoundFloat(val float64, precision uint) float64 {
+//     ratio := math.Pow(10, float64(precision))
+//     return math.Round(val*ratio) / ratio
+// }
+
+// ReadStringFile returns trimmed file contents or empty string on error.
+func ReadStringFile(path string) string {
+	content, _ := ReadStringFileOK(path)
+	return content
+}
+
+// ReadStringFileOK returns trimmed file contents and read success.
+func ReadStringFileOK(path string) (string, bool) {
+	b, err := os.ReadFile(path)
+	if err != nil {
+		return "", false
+	}
+	return strings.TrimSpace(string(b)), true
+}
+
+// ReadStringFileLimited reads a file into a string with a maximum size (in bytes) to avoid
+// allocating large buffers and potential panics with pseudo-files when the size is misreported.
+func ReadStringFileLimited(path string, maxSize int) (string, error) {
+	f, err := os.Open(path)
+	if err != nil {
+		return "", err
+	}
+	defer f.Close()
+
+	buf := make([]byte, maxSize)
+	n, err := f.Read(buf)
+	if err != nil && err != io.EOF {
+		return "", err
+	}
+	if n < 0 {
+		return "", fmt.Errorf("%s returned negative bytes: %d", path, n)
+	}
+	return strings.TrimSpace(string(buf[:n])), nil
+}
+
+// FileExists reports whether the given path exists.
+func FileExists(path string) bool {
+	_, err := os.Stat(path)
+	return err == nil
+}
+
+// ReadUintFile parses a decimal uint64 value from a file.
+func ReadUintFile(path string) (uint64, bool) {
+	raw, ok := ReadStringFileOK(path)
+	if !ok {
+		return 0, false
+	}
+	parsed, err := strconv.ParseUint(raw, 10, 64)
+	if err != nil {
+		return 0, false
+	}
+	return parsed, true
+}
+
+// LookPathHomebrew is like exec.LookPath but also checks Homebrew paths.
+func LookPathHomebrew(file string) (string, error) {
+	foundPath, lookPathErr := exec.LookPath(file)
+	if lookPathErr == nil {
+		return foundPath, nil
+	}
+	var homebrewPath string
+	switch runtime.GOOS {
+	case "darwin":
+		homebrewPath = filepath.Join("/opt", "homebrew", "bin", file)
+	case "linux":
+		homebrewPath = filepath.Join("/home", "linuxbrew", ".linuxbrew", "bin", file)
+	}
+	if homebrewPath != "" {
+		if _, err := os.Stat(homebrewPath); err == nil {
+			return homebrewPath, nil
+		}
+	}
+	return "", lookPathErr
+}

agent/utils/utils_test.go

@@ -0,0 +1,158 @@
+package utils
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestTwoDecimals(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    float64
+		expected float64
+	}{
+		{"round down", 1.234, 1.23},
+		{"round half up", 1.235, 1.24}, // math.Round rounds half up
+		{"no rounding needed", 1.23, 1.23},
+		{"negative number", -1.235, -1.24}, // math.Round rounds half up (more negative)
+		{"zero", 0.0, 0.0},
+		{"large number", 123.456, 123.46}, // rounds 5 up
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := TwoDecimals(tt.input)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}
+
+func TestBytesToMegabytes(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    float64
+		expected float64
+	}{
+		{"1 MB", 1048576, 1.0},
+		{"512 KB", 524288, 0.5},
+		{"zero", 0, 0},
+		{"large value", 1073741824, 1024}, // 1 GB = 1024 MB
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := BytesToMegabytes(tt.input)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}
+
+func TestBytesToGigabytes(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    uint64
+		expected float64
+	}{
+		{"1 GB", 1073741824, 1.0},
+		{"512 MB", 536870912, 0.5},
+		{"0 GB", 0, 0},
+		{"2 GB", 2147483648, 2.0},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := BytesToGigabytes(tt.input)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}
+
+func TestFileFunctions(t *testing.T) {
+	tmpDir := t.TempDir()
+	testFilePath := filepath.Join(tmpDir, "test.txt")
+	testContent := "hello world"
+
+	// Test FileExists (false)
+	assert.False(t, FileExists(testFilePath))
+
+	// Test ReadStringFileOK (false)
+	content, ok := ReadStringFileOK(testFilePath)
+	assert.False(t, ok)
+	assert.Empty(t, content)
+
+	// Test ReadStringFile (empty)
+	assert.Empty(t, ReadStringFile(testFilePath))
+
+	// Write file
+	err := os.WriteFile(testFilePath, []byte(testContent+"\n "), 0644)
+	assert.NoError(t, err)
+
+	// Test FileExists (true)
+	assert.True(t, FileExists(testFilePath))
+
+	// Test ReadStringFileOK (true)
+	content, ok = ReadStringFileOK(testFilePath)
+	assert.True(t, ok)
+	assert.Equal(t, testContent, content)
+
+	// Test ReadStringFile (content)
+	assert.Equal(t, testContent, ReadStringFile(testFilePath))
+}
+
+func TestReadUintFile(t *testing.T) {
+	tmpDir := t.TempDir()
+
+	t.Run("valid uint", func(t *testing.T) {
+		path := filepath.Join(tmpDir, "uint.txt")
+		os.WriteFile(path, []byte(" 12345\n"), 0644)
+		val, ok := ReadUintFile(path)
+		assert.True(t, ok)
+		assert.Equal(t, uint64(12345), val)
+	})
+
+	t.Run("invalid uint", func(t *testing.T) {
+		path := filepath.Join(tmpDir, "invalid.txt")
+		os.WriteFile(path, []byte("abc"), 0644)
+		val, ok := ReadUintFile(path)
+		assert.False(t, ok)
+		assert.Equal(t, uint64(0), val)
+	})
+
+	t.Run("missing file", func(t *testing.T) {
+		path := filepath.Join(tmpDir, "missing.txt")
+		val, ok := ReadUintFile(path)
+		assert.False(t, ok)
+		assert.Equal(t, uint64(0), val)
+	})
+}
+
+func TestGetEnv(t *testing.T) {
+	key := "TEST_VAR"
+	prefixedKey := "BESZEL_AGENT_" + key
+
+	t.Run("prefixed variable exists", func(t *testing.T) {
+		t.Setenv(prefixedKey, "prefixed_val")
+		t.Setenv(key, "unprefixed_val")
+
+		val, exists := GetEnv(key)
+		assert.True(t, exists)
+		assert.Equal(t, "prefixed_val", val)
+	})
+
+	t.Run("only unprefixed variable exists", func(t *testing.T) {
+		t.Setenv(key, "unprefixed_val")
+
+		val, exists := GetEnv(key)
+		assert.True(t, exists)
+		assert.Equal(t, "unprefixed_val", val)
+	})
+
+	t.Run("neither variable exists", func(t *testing.T) {
+		val, exists := GetEnv(key)
+		assert.False(t, exists)
+		assert.Empty(t, val)
+	})
+}

agent/zfs/zfs_freebsd.go

@@ -0,0 +1,11 @@
+//go:build freebsd
+
+package zfs
+
+import (
+	"golang.org/x/sys/unix"
+)
+
+func ARCSize() (uint64, error) {
+	return unix.SysctlUint64("kstat.zfs.misc.arcstats.size")
+}

agent/zfs/zfs_linux.go

@@ -0,0 +1,34 @@
+//go:build linux
+
+// Package zfs provides functions to read ZFS statistics.
+package zfs
+
+import (
+	"bufio"
+	"fmt"
+	"os"
+	"strconv"
+	"strings"
+)
+
+func ARCSize() (uint64, error) {
+	file, err := os.Open("/proc/spl/kstat/zfs/arcstats")
+	if err != nil {
+		return 0, err
+	}
+	defer file.Close()
+
+	scanner := bufio.NewScanner(file)
+	for scanner.Scan() {
+		line := scanner.Text()
+		if strings.HasPrefix(line, "size") {
+			fields := strings.Fields(line)
+			if len(fields) < 3 {
+				return 0, fmt.Errorf("unexpected arcstats size format: %s", line)
+			}
+			return strconv.ParseUint(fields[2], 10, 64)
+		}
+	}
+
+	return 0, fmt.Errorf("size field not found in arcstats")
+}

agent/zfs/zfs_unsupported.go

@@ -0,0 +1,9 @@
+//go:build !linux && !freebsd
+
+package zfs
+
+import "errors"
+
+func ARCSize() (uint64, error) {
+	return 0, errors.ErrUnsupported
+}

beszel.go

@@ -6,7 +6,7 @@ import "github.com/blang/semver"
 
 const (
 	// Version is the current version of the application.
-	Version = "0.18.3"
+	Version = "0.18.7"
 	// AppName is the name of the application.
 	AppName = "beszel"
 )

go.mod

@@ -1,28 +1,28 @@
 module github.com/henrygd/beszel
 
-go 1.25.5
+go 1.26.1
 
 require (
 	github.com/blang/semver v3.5.1+incompatible
 	github.com/coreos/go-systemd/v22 v22.7.0
-	github.com/distatus/battery v0.11.0
-	github.com/ebitengine/purego v0.9.1
+	github.com/ebitengine/purego v0.10.0
 	github.com/fxamacker/cbor/v2 v2.9.0
 	github.com/gliderlabs/ssh v0.3.8
 	github.com/google/uuid v1.6.0
-	github.com/lxzan/gws v1.8.9
-	github.com/nicholas-fedor/shoutrrr v0.13.1
-	github.com/pocketbase/dbx v1.11.0
-	github.com/pocketbase/pocketbase v0.36.2
-	github.com/shirou/gopsutil/v4 v4.26.1
+	github.com/lxzan/gws v1.9.1
+	github.com/nicholas-fedor/shoutrrr v0.14.3
+	github.com/pocketbase/dbx v1.12.0
+	github.com/pocketbase/pocketbase v0.36.8
+	github.com/shirou/gopsutil/v4 v4.26.3
 	github.com/spf13/cast v1.10.0
 	github.com/spf13/cobra v1.10.2
 	github.com/spf13/pflag v1.0.10
 	github.com/stretchr/testify v1.11.1
-	golang.org/x/crypto v0.47.0
-	golang.org/x/exp v0.0.0-20260112195511-716be5621a96
-	golang.org/x/sys v0.40.0
+	golang.org/x/crypto v0.49.0
+	golang.org/x/exp v0.0.0-20260312153236-7ab1446f8b90
+	golang.org/x/sys v0.42.0
 	gopkg.in/yaml.v3 v3.0.1
+	howett.net/plist v1.0.1
 )
 
 require (
@@ -30,10 +30,10 @@ require (
 	github.com/asaskevich/govalidator v0.0.0-20230301143203-a9d515a09cc2 // indirect
 	github.com/davecgh/go-spew v1.1.2-0.20180830191138-d8f796af33cc // indirect
 	github.com/disintegration/imaging v1.6.2 // indirect
-	github.com/dolthub/maphash v0.1.0 // indirect
 	github.com/domodwyer/mailyak/v3 v3.6.2 // indirect
 	github.com/dustin/go-humanize v1.0.1 // indirect
-	github.com/fatih/color v1.18.0 // indirect
+	github.com/eclipse/paho.golang v0.23.0 // indirect
+	github.com/fatih/color v1.19.0 // indirect
 	github.com/gabriel-vasile/mimetype v1.4.13 // indirect
 	github.com/ganigeorgiev/fexpr v0.5.0 // indirect
 	github.com/go-ole/go-ole v1.3.0 // indirect
@@ -41,9 +41,10 @@ require (
 	github.com/go-sql-driver/mysql v1.9.1 // indirect
 	github.com/godbus/dbus/v5 v5.2.2 // indirect
 	github.com/golang-jwt/jwt/v5 v5.3.1 // indirect
+	github.com/gorilla/websocket v1.5.3 // indirect
 	github.com/inconshreveable/mousetrap v1.1.0 // indirect
-	github.com/klauspost/compress v1.18.3 // indirect
-	github.com/lufia/plan9stats v0.0.0-20251013123823-9fd1530e3ec3 // indirect
+	github.com/klauspost/compress v1.18.5 // indirect
+	github.com/lufia/plan9stats v0.0.0-20260324052639-156f7da3f749 // indirect
 	github.com/mattn/go-colorable v0.1.14 // indirect
 	github.com/mattn/go-isatty v0.0.20 // indirect
 	github.com/ncruces/go-strftime v1.0.0 // indirect
@@ -54,15 +55,14 @@ require (
 	github.com/tklauser/numcpus v0.11.0 // indirect
 	github.com/x448/float16 v0.8.4 // indirect
 	github.com/yusufpapurcu/wmi v1.2.4 // indirect
-	golang.org/x/image v0.35.0 // indirect
-	golang.org/x/net v0.49.0 // indirect
-	golang.org/x/oauth2 v0.34.0 // indirect
-	golang.org/x/sync v0.19.0 // indirect
-	golang.org/x/term v0.39.0 // indirect
-	golang.org/x/text v0.33.0 // indirect
-	howett.net/plist v1.0.1 // indirect
-	modernc.org/libc v1.67.6 // indirect
+	golang.org/x/image v0.38.0 // indirect
+	golang.org/x/net v0.52.0 // indirect
+	golang.org/x/oauth2 v0.36.0 // indirect
+	golang.org/x/sync v0.20.0 // indirect
+	golang.org/x/term v0.41.0 // indirect
+	golang.org/x/text v0.35.0 // indirect
+	modernc.org/libc v1.70.0 // indirect
 	modernc.org/mathutil v1.7.1 // indirect
 	modernc.org/memory v1.11.0 // indirect
-	modernc.org/sqlite v1.44.3 // indirect
+	modernc.org/sqlite v1.48.0 // indirect
 )

go.sum

@@ -17,18 +17,16 @@ github.com/davecgh/go-spew v1.1.2-0.20180830191138-d8f796af33cc h1:U9qPSI2PIWSS1
 github.com/davecgh/go-spew v1.1.2-0.20180830191138-d8f796af33cc/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/disintegration/imaging v1.6.2 h1:w1LecBlG2Lnp8B3jk5zSuNqd7b4DXhcjwek1ei82L+c=
 github.com/disintegration/imaging v1.6.2/go.mod h1:44/5580QXChDfwIclfc/PCwrr44amcmDAg8hxG0Ewe4=
-github.com/distatus/battery v0.11.0 h1:KJk89gz90Iq/wJtbjjM9yUzBXV+ASV/EG2WOOL7N8lc=
-github.com/distatus/battery v0.11.0/go.mod h1:KmVkE8A8hpIX4T78QRdMktYpEp35QfOL8A8dwZBxq2k=
-github.com/dolthub/maphash v0.1.0 h1:bsQ7JsF4FkkWyrP3oCnFJgrCUAFbFf3kOl4L/QxPDyQ=
-github.com/dolthub/maphash v0.1.0/go.mod h1:gkg4Ch4CdCDu5h6PMriVLawB7koZ+5ijb9puGMV50a4=
 github.com/domodwyer/mailyak/v3 v3.6.2 h1:x3tGMsyFhTCaxp6ycgR0FE/bu5QiNp+hetUuCOBXMn8=
 github.com/domodwyer/mailyak/v3 v3.6.2/go.mod h1:lOm/u9CyCVWHeaAmHIdF4RiKVxKUT/H5XX10lIKAL6c=
 github.com/dustin/go-humanize v1.0.1 h1:GzkhY7T5VNhEkwH0PVJgjz+fX1rhBrR7pRT3mDkpeCY=
 github.com/dustin/go-humanize v1.0.1/go.mod h1:Mu1zIs6XwVuF/gI1OepvI0qD18qycQx+mFykh5fBlto=
-github.com/ebitengine/purego v0.9.1 h1:a/k2f2HQU3Pi399RPW1MOaZyhKJL9w/xFpKAg4q1s0A=
-github.com/ebitengine/purego v0.9.1/go.mod h1:iIjxzd6CiRiOG0UyXP+V1+jWqUXVjPKLAI0mRfJZTmQ=
-github.com/fatih/color v1.18.0 h1:S8gINlzdQ840/4pfAwic/ZE0djQEH3wM94VfqLTZcOM=
-github.com/fatih/color v1.18.0/go.mod h1:4FelSpRwEGDpQ12mAdzqdOukCy4u8WUtOY6lkT/6HfU=
+github.com/ebitengine/purego v0.10.0 h1:QIw4xfpWT6GWTzaW5XEKy3HXoqrJGx1ijYHzTF0/ISU=
+github.com/ebitengine/purego v0.10.0/go.mod h1:iIjxzd6CiRiOG0UyXP+V1+jWqUXVjPKLAI0mRfJZTmQ=
+github.com/eclipse/paho.golang v0.23.0 h1:KHgl2wz6EJo7cMBmkuhpt7C576vP+kpPv7jjvSyR6Mk=
+github.com/eclipse/paho.golang v0.23.0/go.mod h1:nQRhTkoZv8EAiNs5UU0/WdQIx2NrnWUpL9nsGJTQN04=
+github.com/fatih/color v1.19.0 h1:Zp3PiM21/9Ld6FzSKyL5c/BULoe/ONr9KlbYVOfG8+w=
+github.com/fatih/color v1.19.0/go.mod h1:zNk67I0ZUT1bEGsSGyCZYZNrHuTkJJB+r6Q9VuMi0LE=
 github.com/frankban/quicktest v1.14.6 h1:7Xjx+VpznH+oBnejlPUj8oUpdxnVs4f8XU8WnHkI4W8=
 github.com/frankban/quicktest v1.14.6/go.mod h1:4ptaffx2x8+WTWXmUCuVU6aPUX1/Mz7zb5vbUoiM6w0=
 github.com/fxamacker/cbor/v2 v2.9.0 h1:NpKPmjDBgUfBms6tr6JZkTHtfFGcMKsw3eGcmD/sapM=
@@ -58,10 +56,12 @@ github.com/golang-jwt/jwt/v5 v5.3.1/go.mod h1:fxCRLWMO43lRc8nhHWY6LGqRcf+1gQWArs
 github.com/golang/protobuf v1.3.1/go.mod h1:6lQm79b+lXiMfvg/cZm0SGofjICqVBUtrP5yJMmIC1U=
 github.com/google/go-cmp v0.7.0 h1:wk8382ETsv4JYUZwIsn6YpYiWiBsYLSJiTsyBybVuN8=
 github.com/google/go-cmp v0.7.0/go.mod h1:pXiqmnSA92OHEEa9HXL2W4E7lf9JzCmGVUdgjX3N/iU=
-github.com/google/pprof v0.0.0-20260115054156-294ebfa9ad83 h1:z2ogiKUYzX5Is6zr/vP9vJGqPwcdqsWjOt+V8J7+bTc=
-github.com/google/pprof v0.0.0-20260115054156-294ebfa9ad83/go.mod h1:MxpfABSjhmINe3F1It9d+8exIHFvUqtLIRCdOGNXqiI=
+github.com/google/pprof v0.0.0-20260302011040-a15ffb7f9dcc h1:VBbFa1lDYWEeV5FZKUiYKYT0VxCp9twUmmaq9eb8sXw=
+github.com/google/pprof v0.0.0-20260302011040-a15ffb7f9dcc/go.mod h1:MxpfABSjhmINe3F1It9d+8exIHFvUqtLIRCdOGNXqiI=
 github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
 github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
+github.com/gorilla/websocket v1.5.3 h1:saDtZ6Pbx/0u+bgYQ3q96pZgCzfhKXGPqt7kZ72aNNg=
+github.com/gorilla/websocket v1.5.3/go.mod h1:YR8l580nyteQvAITg2hZ9XVh4b55+EU/adAjf1fMHhE=
 github.com/hashicorp/golang-lru/v2 v2.0.7 h1:a+bsQ5rvGLjzHuww6tVxozPZFVghXaHOwFs4luLUK2k=
 github.com/hashicorp/golang-lru/v2 v2.0.7/go.mod h1:QeFd9opnmA6QUJc5vARoKUSoFhyfM2/ZepoAG6RGpeM=
 github.com/inconshreveable/mousetrap v1.1.0 h1:wN+x4NVGpMsO7ErUn/mUI3vEoE6Jt13X2s0bqwp9tc8=
@@ -69,35 +69,35 @@ github.com/inconshreveable/mousetrap v1.1.0/go.mod h1:vpF70FUmC8bwa3OWnCshd2FqLf
 github.com/jarcoal/httpmock v1.4.1 h1:0Ju+VCFuARfFlhVXFc2HxlcQkfB+Xq12/EotHko+x2A=
 github.com/jarcoal/httpmock v1.4.1/go.mod h1:ftW1xULwo+j0R0JJkJIIi7UKigZUXCLLanykgjwBXL0=
 github.com/jessevdk/go-flags v1.4.0/go.mod h1:4FA24M0QyGHXBuZZK/XkWh8h0e1EYbRYJSGM75WSRxI=
-github.com/klauspost/compress v1.18.3 h1:9PJRvfbmTabkOX8moIpXPbMMbYN60bWImDDU7L+/6zw=
-github.com/klauspost/compress v1.18.3/go.mod h1:R0h/fSBs8DE4ENlcrlib3PsXS61voFxhIs2DeRhCvJ4=
+github.com/klauspost/compress v1.18.5 h1:/h1gH5Ce+VWNLSWqPzOVn6XBO+vJbCNGvjoaGBFW2IE=
+github.com/klauspost/compress v1.18.5/go.mod h1:cwPg85FWrGar70rWktvGQj8/hthj3wpl0PGDogxkrSQ=
 github.com/kr/pretty v0.3.1 h1:flRD4NNwYAUpkphVc1HcthR4KEIFJ65n8Mw5qdRn3LE=
 github.com/kr/pretty v0.3.1/go.mod h1:hoEshYVHaxMs3cyo3Yncou5ZscifuDolrwPKZanG3xk=
 github.com/kr/text v0.2.0 h1:5Nx0Ya0ZqY2ygV366QzturHI13Jq95ApcVaJBhpS+AY=
 github.com/kr/text v0.2.0/go.mod h1:eLer722TekiGuMkidMxC/pM04lWEeraHUUmBw8l2grE=
-github.com/lufia/plan9stats v0.0.0-20251013123823-9fd1530e3ec3 h1:PwQumkgq4/acIiZhtifTV5OUqqiP82UAl0h87xj/l9k=
-github.com/lufia/plan9stats v0.0.0-20251013123823-9fd1530e3ec3/go.mod h1:autxFIvghDt3jPTLoqZ9OZ7s9qTGNAWmYCjVFWPX/zg=
-github.com/lxzan/gws v1.8.9 h1:VU3SGUeWlQrEwfUSfokcZep8mdg/BrUF+y73YYshdBM=
-github.com/lxzan/gws v1.8.9/go.mod h1:d9yHaR1eDTBHagQC6KY7ycUOaz5KWeqQtP3xu7aMK8Y=
+github.com/lufia/plan9stats v0.0.0-20260324052639-156f7da3f749 h1:Qj3hTcdWH8uMZDI41HNuTuJN525C7NBrbtH5kSO6fPk=
+github.com/lufia/plan9stats v0.0.0-20260324052639-156f7da3f749/go.mod h1:autxFIvghDt3jPTLoqZ9OZ7s9qTGNAWmYCjVFWPX/zg=
+github.com/lxzan/gws v1.9.1 h1:4lbIp4cW0hOLP3ejFHR/uWRy741AURx7oKkNNi2OT9o=
+github.com/lxzan/gws v1.9.1/go.mod h1:gXHSCPmTGryWJ4icuqy8Yho32E4YIMHH0fkDRYJRbdc=
 github.com/mattn/go-colorable v0.1.14 h1:9A9LHSqF/7dyVVX6g0U9cwm9pG3kP9gSzcuIPHPsaIE=
 github.com/mattn/go-colorable v0.1.14/go.mod h1:6LmQG8QLFO4G5z1gPvYEzlUgJ2wF+stgPZH1UqBm1s8=
 github.com/mattn/go-isatty v0.0.20 h1:xfD0iDuEKnDkl03q4limB+vH+GxLEtL/jb4xVJSWWEY=
 github.com/mattn/go-isatty v0.0.20/go.mod h1:W+V8PltTTMOvKvAeJH7IuucS94S2C6jfK/D7dTCTo3Y=
 github.com/ncruces/go-strftime v1.0.0 h1:HMFp8mLCTPp341M/ZnA4qaf7ZlsbTc+miZjCLOFAw7w=
 github.com/ncruces/go-strftime v1.0.0/go.mod h1:Fwc5htZGVVkseilnfgOVb9mKy6w1naJmn9CehxcKcls=
-github.com/nicholas-fedor/shoutrrr v0.13.1 h1:llEoHNbnMM4GfQ9+2Ns3n6ssvNfi3NPWluM0AQiicoY=
-github.com/nicholas-fedor/shoutrrr v0.13.1/go.mod h1:kU4cFJpEAtTzl3iV0l+XUXmM90OlC5T01b7roM4/pYM=
-github.com/onsi/ginkgo/v2 v2.27.3 h1:ICsZJ8JoYafeXFFlFAG75a7CxMsJHwgKwtO+82SE9L8=
-github.com/onsi/ginkgo/v2 v2.27.3/go.mod h1:ArE1D/XhNXBXCBkKOLkbsb2c81dQHCRcF5zwn/ykDRo=
-github.com/onsi/gomega v1.38.3 h1:eTX+W6dobAYfFeGC2PV6RwXRu/MyT+cQguijutvkpSM=
-github.com/onsi/gomega v1.38.3/go.mod h1:ZCU1pkQcXDO5Sl9/VVEGlDyp+zm0m1cmeG5TOzLgdh4=
+github.com/nicholas-fedor/shoutrrr v0.14.3 h1:aBX2iw9a7jl5wfHd3bi9LnS5ucoYIy6KcLH9XVF+gig=
+github.com/nicholas-fedor/shoutrrr v0.14.3/go.mod h1:U7IywBkLpBV7rgn8iLbQ9/LklJG1gm24bFv5cXXsDKs=
+github.com/onsi/ginkgo/v2 v2.28.1 h1:S4hj+HbZp40fNKuLUQOYLDgZLwNUVn19N3Atb98NCyI=
+github.com/onsi/ginkgo/v2 v2.28.1/go.mod h1:CLtbVInNckU3/+gC8LzkGUb9oF+e8W8TdUsxPwvdOgE=
+github.com/onsi/gomega v1.39.1 h1:1IJLAad4zjPn2PsnhH70V4DKRFlrCzGBNrNaru+Vf28=
+github.com/onsi/gomega v1.39.1/go.mod h1:hL6yVALoTOxeWudERyfppUcZXjMwIMLnuSfruD2lcfg=
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
 github.com/pmezard/go-difflib v1.0.1-0.20181226105442-5d4384ee4fb2 h1:Jamvg5psRIccs7FGNTlIRMkT8wgtp5eCXdBlqhYGL6U=
 github.com/pmezard/go-difflib v1.0.1-0.20181226105442-5d4384ee4fb2/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
-github.com/pocketbase/dbx v1.11.0 h1:LpZezioMfT3K4tLrqA55wWFw1EtH1pM4tzSVa7kgszU=
-github.com/pocketbase/dbx v1.11.0/go.mod h1:xXRCIAKTHMgUCyCKZm55pUOdvFziJjQfXaWKhu2vhMs=
-github.com/pocketbase/pocketbase v0.36.2 h1:mzrxnvXKc3yxKlvZdbwoYXkH8kfIETteD0hWdgj0VI4=
-github.com/pocketbase/pocketbase v0.36.2/go.mod h1:71vSF8whUDzC8mcLFE10+Qatf9JQdeOGIRWawOuLLKM=
+github.com/pocketbase/dbx v1.12.0 h1:/oLErM+A0b4xI0PWTGPqSDVjzix48PqI/bng2l0PzoA=
+github.com/pocketbase/dbx v1.12.0/go.mod h1:xXRCIAKTHMgUCyCKZm55pUOdvFziJjQfXaWKhu2vhMs=
+github.com/pocketbase/pocketbase v0.36.8 h1:gCNqoesZ44saYOD3J7edhi5nDwUWKyQG7boM/kVwz2c=
+github.com/pocketbase/pocketbase v0.36.8/go.mod h1:OY4WaXbP0WnF/EXoBbboWJK+ZSZ1A85tiA0sjrTKxTA=
 github.com/power-devops/perfstat v0.0.0-20240221224432-82ca36839d55 h1:o4JXh1EVt9k/+g42oCprj/FisM4qX9L3sZB3upGN2ZU=
 github.com/power-devops/perfstat v0.0.0-20240221224432-82ca36839d55/go.mod h1:OmDBASR4679mdNQnz2pUhc2G8CO2JrUAVFDRBDP/hJE=
 github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec h1:W09IVJc94icq4NjY3clb7Lk8O1qJ8BdBEF8z0ibU0rE=
@@ -105,8 +105,8 @@ github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec/go.mod h1:qq
 github.com/rogpeppe/go-internal v1.9.0 h1:73kH8U+JUqXU8lRuOHeVHaa/SZPifC7BkcraZVejAe8=
 github.com/rogpeppe/go-internal v1.9.0/go.mod h1:WtVeX8xhTBvf0smdhujwtBcq4Qrzq/fJaraNFVN+nFs=
 github.com/russross/blackfriday/v2 v2.1.0/go.mod h1:+Rmxgy9KzJVeS9/2gXHxylqXiyQDYRxCVz55jmeOWTM=
-github.com/shirou/gopsutil/v4 v4.26.1 h1:TOkEyriIXk2HX9d4isZJtbjXbEjf5qyKPAzbzY0JWSo=
-github.com/shirou/gopsutil/v4 v4.26.1/go.mod h1:medLI9/UNAb0dOI9Q3/7yWSqKkj00u+1tgY8nvv41pc=
+github.com/shirou/gopsutil/v4 v4.26.3 h1:2ESdQt90yU3oXF/CdOlRCJxrP+Am1aBYubTMTfxJ1qc=
+github.com/shirou/gopsutil/v4 v4.26.3/go.mod h1:LZ6ewCSkBqUpvSOf+LsTGnRinC6iaNUNMGBtDkJBaLQ=
 github.com/spf13/cast v1.10.0 h1:h2x0u2shc1QuLHfxi+cTJvs30+ZAHOGRic8uyGTDWxY=
 github.com/spf13/cast v1.10.0/go.mod h1:jNfB8QC9IA6ZuY2ZjDp0KtFO2LZZlg4S/7bzP6qqeHo=
 github.com/spf13/cobra v1.10.2 h1:DMTTonx5m65Ic0GOoRY2c16WCbHxOOw6xxezuLaBpcU=
@@ -115,6 +115,8 @@ github.com/spf13/pflag v1.0.9/go.mod h1:McXfInJRrz4CZXVZOBLb0bTZqETkiAhM9Iw0y3An
 github.com/spf13/pflag v1.0.10 h1:4EBh2KAYBwaONj6b2Ye1GiHfwjqyROoF4RwYO+vPwFk=
 github.com/spf13/pflag v1.0.10/go.mod h1:McXfInJRrz4CZXVZOBLb0bTZqETkiAhM9Iw0y3An2Bg=
 github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
+github.com/stretchr/objx v0.5.3 h1:jmXUvGomnU1o3W/V5h2VEradbpJDwGrzugQQvL0POH4=
+github.com/stretchr/objx v0.5.3/go.mod h1:rDQraq+vQZU7Fde9LOZLr8Tax6zZvy4kuNKF+QYS+U0=
 github.com/stretchr/testify v1.4.0/go.mod h1:j7eGeouHqKxXV5pUuKE4zz7dFj8WfuZ+81PSLYec5m4=
 github.com/stretchr/testify v1.11.1 h1:7s2iGBzp5EwR7/aIZr8ao5+dra3wiQyKjjFuvgVKu7U=
 github.com/stretchr/testify v1.11.1/go.mod h1:wZwfW3scLgRK+23gO65QZefKpKQRnfz6sD981Nm4B6U=
@@ -126,44 +128,44 @@ github.com/x448/float16 v0.8.4 h1:qLwI1I70+NjRFUR3zs1JPUCgaCXSh3SW62uAKT1mSBM=
 github.com/x448/float16 v0.8.4/go.mod h1:14CWIYCyZA/cWjXOioeEpHeN/83MdbZDRQHoFcYsOfg=
 github.com/yusufpapurcu/wmi v1.2.4 h1:zFUKzehAFReQwLys1b/iSMl+JQGSCSjtVqQn9bBrPo0=
 github.com/yusufpapurcu/wmi v1.2.4/go.mod h1:SBZ9tNy3G9/m5Oi98Zks0QjeHVDvuK0qfxQmPyzfmi0=
+go.uber.org/goleak v1.3.0 h1:2K3zAYmnTNqV73imy9J1T3WC+gmCePx2hEGkimedGto=
+go.uber.org/goleak v1.3.0/go.mod h1:CoHD4mav9JJNrW/WLlf7HGZPjdw8EucARQHekz1X6bE=
 go.yaml.in/yaml/v3 v3.0.4 h1:tfq32ie2Jv2UxXFdLJdh3jXuOzWiL1fo0bu/FbuKpbc=
 go.yaml.in/yaml/v3 v3.0.4/go.mod h1:DhzuOOF2ATzADvBadXxruRBLzYTpT36CKvDb3+aBEFg=
 golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
-golang.org/x/crypto v0.47.0 h1:V6e3FRj+n4dbpw86FJ8Fv7XVOql7TEwpHapKoMJ/GO8=
-golang.org/x/crypto v0.47.0/go.mod h1:ff3Y9VzzKbwSSEzWqJsJVBnWmRwRSHt/6Op5n9bQc4A=
-golang.org/x/exp v0.0.0-20260112195511-716be5621a96 h1:Z/6YuSHTLOHfNFdb8zVZomZr7cqNgTJvA8+Qz75D8gU=
-golang.org/x/exp v0.0.0-20260112195511-716be5621a96/go.mod h1:nzimsREAkjBCIEFtHiYkrJyT+2uy9YZJB7H1k68CXZU=
+golang.org/x/crypto v0.49.0 h1:+Ng2ULVvLHnJ/ZFEq4KdcDd/cfjrrjjNSXNzxg0Y4U4=
+golang.org/x/crypto v0.49.0/go.mod h1:ErX4dUh2UM+CFYiXZRTcMpEcN8b/1gxEuv3nODoYtCA=
+golang.org/x/exp v0.0.0-20260312153236-7ab1446f8b90 h1:jiDhWWeC7jfWqR9c/uplMOqJ0sbNlNWv0UkzE0vX1MA=
+golang.org/x/exp v0.0.0-20260312153236-7ab1446f8b90/go.mod h1:xE1HEv6b+1SCZ5/uscMRjUBKtIxworgEcEi+/n9NQDQ=
 golang.org/x/image v0.0.0-20191009234506-e7c1f5e7dbb8/go.mod h1:FeLwcggjj3mMvU+oOTbSwawSJRM1uh48EjtB4UJZlP0=
-golang.org/x/image v0.35.0 h1:LKjiHdgMtO8z7Fh18nGY6KDcoEtVfsgLDPeLyguqb7I=
-golang.org/x/image v0.35.0/go.mod h1:MwPLTVgvxSASsxdLzKrl8BRFuyqMyGhLwmC+TO1Sybk=
-golang.org/x/mod v0.32.0 h1:9F4d3PHLljb6x//jOyokMv3eX+YDeepZSEo3mFJy93c=
-golang.org/x/mod v0.32.0/go.mod h1:SgipZ/3h2Ci89DlEtEXWUk/HteuRin+HHhN+WbNhguU=
+golang.org/x/image v0.38.0 h1:5l+q+Y9JDC7mBOMjo4/aPhMDcxEptsX+Tt3GgRQRPuE=
+golang.org/x/image v0.38.0/go.mod h1:/3f6vaXC+6CEanU4KJxbcUZyEePbyKbaLoDOe4ehFYY=
+golang.org/x/mod v0.34.0 h1:xIHgNUUnW6sYkcM5Jleh05DvLOtwc6RitGHbDk4akRI=
+golang.org/x/mod v0.34.0/go.mod h1:ykgH52iCZe79kzLLMhyCUzhMci+nQj+0XkbXpNYtVjY=
 golang.org/x/net v0.0.0-20190603091049-60506f45cf65/go.mod h1:HSz+uSET+XFnRR8LxR5pz3Of3rY3CfYBVs4xY44aLks=
-golang.org/x/net v0.49.0 h1:eeHFmOGUTtaaPSGNmjBKpbng9MulQsJURQUAfUwY++o=
-golang.org/x/net v0.49.0/go.mod h1:/ysNB2EvaqvesRkuLAyjI1ycPZlQHM3q01F02UY/MV8=
-golang.org/x/oauth2 v0.34.0 h1:hqK/t4AKgbqWkdkcAeI8XLmbK+4m4G5YeQRrmiotGlw=
-golang.org/x/oauth2 v0.34.0/go.mod h1:lzm5WQJQwKZ3nwavOZ3IS5Aulzxi68dUSgRHujetwEA=
-golang.org/x/sync v0.19.0 h1:vV+1eWNmZ5geRlYjzm2adRgW2/mcpevXNg50YZtPCE4=
-golang.org/x/sync v0.19.0/go.mod h1:9KTHXmSnoGruLpwFjVSX0lNNA75CykiMECbovNTZqGI=
+golang.org/x/net v0.52.0 h1:He/TN1l0e4mmR3QqHMT2Xab3Aj3L9qjbhRm78/6jrW0=
+golang.org/x/net v0.52.0/go.mod h1:R1MAz7uMZxVMualyPXb+VaqGSa3LIaUqk0eEt3w36Sw=
+golang.org/x/oauth2 v0.36.0 h1:peZ/1z27fi9hUOFCAZaHyrpWG5lwe0RJEEEeH0ThlIs=
+golang.org/x/oauth2 v0.36.0/go.mod h1:YDBUJMTkDnJS+A4BP4eZBjCqtokkg1hODuPjwiGPO7Q=
+golang.org/x/sync v0.20.0 h1:e0PTpb7pjO8GAtTs2dQ6jYa5BWYlMuX047Dco/pItO4=
+golang.org/x/sync v0.20.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
 golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
 golang.org/x/sys v0.0.0-20190916202348-b4ddaad3f8a3/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20201204225414-ed752295db88/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.1.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.40.0 h1:DBZZqJ2Rkml6QMQsZywtnjnnGvHza6BTfYFWY9kjEWQ=
-golang.org/x/sys v0.40.0/go.mod h1:OgkHotnGiDImocRcuBABYBEXf8A9a87e/uXjp9XT3ks=
-golang.org/x/term v0.39.0 h1:RclSuaJf32jOqZz74CkPA9qFuVTX7vhLlpfj/IGWlqY=
-golang.org/x/term v0.39.0/go.mod h1:yxzUCTP/U+FzoxfdKmLaA0RV1WgE0VY7hXBwKtY/4ww=
+golang.org/x/sys v0.42.0 h1:omrd2nAlyT5ESRdCLYdm3+fMfNFE/+Rf4bDIQImRJeo=
+golang.org/x/sys v0.42.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
+golang.org/x/term v0.41.0 h1:QCgPso/Q3RTJx2Th4bDLqML4W6iJiaXFq2/ftQF13YU=
+golang.org/x/term v0.41.0/go.mod h1:3pfBgksrReYfZ5lvYM0kSO0LIkAl4Yl2bXOkKP7Ec2A=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/text v0.3.2/go.mod h1:bEr9sfX3Q8Zfm5fL9x+3itogRgK3+ptLWKqgva+5dAk=
-golang.org/x/text v0.33.0 h1:B3njUFyqtHDUI5jMn1YIr5B0IE2U0qck04r6d4KPAxE=
-golang.org/x/text v0.33.0/go.mod h1:LuMebE6+rBincTi9+xWTY8TztLzKHc/9C1uBCG27+q8=
+golang.org/x/text v0.35.0 h1:JOVx6vVDFokkpaq1AEptVzLTpDe9KGpj5tR4/X+ybL8=
+golang.org/x/text v0.35.0/go.mod h1:khi/HExzZJ2pGnjenulevKNX1W67CUy0AsXcNubPGCA=
 golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
-golang.org/x/tools v0.41.0 h1:a9b8iMweWG+S0OBnlU36rzLp20z1Rp10w+IY2czHTQc=
-golang.org/x/tools v0.41.0/go.mod h1:XSY6eDqxVNiYgezAVqqCeihT4j1U2CCsqvH3WhQpnlg=
+golang.org/x/tools v0.43.0 h1:12BdW9CeB3Z+J/I/wj34VMl8X+fEXBxVR90JeMX5E7s=
+golang.org/x/tools v0.43.0/go.mod h1:uHkMso649BX2cZK6+RpuIPXS3ho2hZo4FVwfoy1vIk0=
 google.golang.org/appengine v1.6.5/go.mod h1:8WjMMxjGQR8xUklV/ARdw2HLXBOI7O7uCIDZVag1xfc=
-google.golang.org/protobuf v1.36.10 h1:AYd7cD/uASjIL6Q9LiTjz8JLcrh/88q5UObnmY3aOOE=
-google.golang.org/protobuf v1.36.10/go.mod h1:HTf+CrKn2C3g5S8VImy6tdcUvCska2kB7j23XfzDpco=
 gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
 gopkg.in/check.v1 v1.0.0-20201130134442-10cb98267c6c h1:Hei/4ADfdWqJk1ZMxUNpqntNwaWcugrBjAiHlqqRiVk=
 gopkg.in/check.v1 v1.0.0-20201130134442-10cb98267c6c/go.mod h1:JHkPIbrfpd72SG/EVd6muEfDQjcINNoR0C8j2r3qZ4Q=
@@ -175,18 +177,18 @@ howett.net/plist v1.0.1 h1:37GdZ8tP09Q35o9ych3ehygcsL+HqKSwzctveSlarvM=
 howett.net/plist v1.0.1/go.mod h1:lqaXoTrLY4hg8tnEzNru53gicrbv7rrk+2xJA/7hw9g=
 modernc.org/cc/v4 v4.27.1 h1:9W30zRlYrefrDV2JE2O8VDtJ1yPGownxciz5rrbQZis=
 modernc.org/cc/v4 v4.27.1/go.mod h1:uVtb5OGqUKpoLWhqwNQo/8LwvoiEBLvZXIQ/SmO6mL0=
-modernc.org/ccgo/v4 v4.30.1 h1:4r4U1J6Fhj98NKfSjnPUN7Ze2c6MnAdL0hWw6+LrJpc=
-modernc.org/ccgo/v4 v4.30.1/go.mod h1:bIOeI1JL54Utlxn+LwrFyjCx2n2RDiYEaJVSrgdrRfM=
-modernc.org/fileutil v1.3.40 h1:ZGMswMNc9JOCrcrakF1HrvmergNLAmxOPjizirpfqBA=
-modernc.org/fileutil v1.3.40/go.mod h1:HxmghZSZVAz/LXcMNwZPA/DRrQZEVP9VX0V4LQGQFOc=
+modernc.org/ccgo/v4 v4.32.0 h1:hjG66bI/kqIPX1b2yT6fr/jt+QedtP2fqojG2VrFuVw=
+modernc.org/ccgo/v4 v4.32.0/go.mod h1:6F08EBCx5uQc38kMGl+0Nm0oWczoo1c7cgpzEry7Uc0=
+modernc.org/fileutil v1.4.0 h1:j6ZzNTftVS054gi281TyLjHPp6CPHr2KCxEXjEbD6SM=
+modernc.org/fileutil v1.4.0/go.mod h1:EqdKFDxiByqxLk8ozOxObDSfcVOv/54xDs/DUHdvCUU=
 modernc.org/gc/v2 v2.6.5 h1:nyqdV8q46KvTpZlsw66kWqwXRHdjIlJOhG6kxiV/9xI=
 modernc.org/gc/v2 v2.6.5/go.mod h1:YgIahr1ypgfe7chRuJi2gD7DBQiKSLMPgBQe9oIiito=
-modernc.org/gc/v3 v3.1.1 h1:k8T3gkXWY9sEiytKhcgyiZ2L0DTyCQ/nvX+LoCljoRE=
-modernc.org/gc/v3 v3.1.1/go.mod h1:HFK/6AGESC7Ex+EZJhJ2Gni6cTaYpSMmU/cT9RmlfYY=
+modernc.org/gc/v3 v3.1.2 h1:ZtDCnhonXSZexk/AYsegNRV1lJGgaNZJuKjJSWKyEqo=
+modernc.org/gc/v3 v3.1.2/go.mod h1:HFK/6AGESC7Ex+EZJhJ2Gni6cTaYpSMmU/cT9RmlfYY=
 modernc.org/goabi0 v0.2.0 h1:HvEowk7LxcPd0eq6mVOAEMai46V+i7Jrj13t4AzuNks=
 modernc.org/goabi0 v0.2.0/go.mod h1:CEFRnnJhKvWT1c1JTI3Avm+tgOWbkOu5oPA8eH8LnMI=
-modernc.org/libc v1.67.6 h1:eVOQvpModVLKOdT+LvBPjdQqfrZq+pC39BygcT+E7OI=
-modernc.org/libc v1.67.6/go.mod h1:JAhxUVlolfYDErnwiqaLvUqc8nfb2r6S6slAgZOnaiE=
+modernc.org/libc v1.70.0 h1:U58NawXqXbgpZ/dcdS9kMshu08aiA6b7gusEusqzNkw=
+modernc.org/libc v1.70.0/go.mod h1:OVmxFGP1CI/Z4L3E0Q3Mf1PDE0BucwMkcXjjLntvHJo=
 modernc.org/mathutil v1.7.1 h1:GCZVGXdaN8gTqB1Mf/usp1Y/hSqgI2vAGGP4jZMCxOU=
 modernc.org/mathutil v1.7.1/go.mod h1:4p5IwJITfppl0G4sUEDtCr4DthTaT47/N3aT6MhfgJg=
 modernc.org/memory v1.11.0 h1:o4QC8aMQzmcwCK3t3Ux/ZHmwFPzE6hf2Y5LbkRs+hbI=
@@ -195,8 +197,8 @@ modernc.org/opt v0.1.4 h1:2kNGMRiUjrp4LcaPuLY2PzUfqM/w9N23quVwhKt5Qm8=
 modernc.org/opt v0.1.4/go.mod h1:03fq9lsNfvkYSfxrfUhZCWPk1lm4cq4N+Bh//bEtgns=
 modernc.org/sortutil v1.2.1 h1:+xyoGf15mM3NMlPDnFqrteY07klSFxLElE2PVuWIJ7w=
 modernc.org/sortutil v1.2.1/go.mod h1:7ZI3a3REbai7gzCLcotuw9AC4VZVpYMjDzETGsSMqJE=
-modernc.org/sqlite v1.44.3 h1:+39JvV/HWMcYslAwRxHb8067w+2zowvFOUrOWIy9PjY=
-modernc.org/sqlite v1.44.3/go.mod h1:CzbrU2lSB1DKUusvwGz7rqEKIq+NUd8GWuBBZDs9/nA=
+modernc.org/sqlite v1.48.0 h1:ElZyLop3Q2mHYk5IFPPXADejZrlHu7APbpB0sF78bq4=
+modernc.org/sqlite v1.48.0/go.mod h1:hWjRO6Tj/5Ik8ieqxQybiEOUXy0NJFNp2tpvVpKlvig=
 modernc.org/strutil v1.2.1 h1:UneZBkQA+DX2Rp35KcM69cSsNES9ly8mQWD71HKlOA0=
 modernc.org/strutil v1.2.1/go.mod h1:EHkiggD70koQxjVdSBM3JKM7k6L0FbGE5eymy9i3B9A=
 modernc.org/token v1.1.0 h1:Xl7Ap9dKaEs5kLoOQeQmPWevfnk/DM5qcLcYlA8ys6Y=

internal/alerts/alerts.go

@@ -21,9 +21,9 @@ type hubLike interface {
 
 type AlertManager struct {
 	hub           hubLike
-	alertQueue    chan alertTask
-	stopChan      chan struct{}
+	stopOnce      sync.Once
 	pendingAlerts sync.Map
+	alertsCache   *AlertsCache
 }
 
 type AlertMessageData struct {
@@ -40,16 +40,22 @@ type UserNotificationSettings struct {
 	Webhooks []string `json:"webhooks"`
 }
 
+type SystemAlertFsStats struct {
+	DiskTotal float64 `json:"d"`
+	DiskUsed  float64 `json:"du"`
+}
+
+// Values pulled from system_stats.stats that are relevant to alerts.
 type SystemAlertStats struct {
 	Cpu          float64                       `json:"cpu"`
 	Mem          float64                       `json:"mp"`
 	Disk         float64                       `json:"dp"`
-	NetSent      float64                       `json:"ns"`
-	NetRecv      float64                       `json:"nr"`
+	Bandwidth    [2]uint64                     `json:"b"`
 	GPU          map[string]SystemAlertGPUData `json:"g"`
 	Temperatures map[string]float32            `json:"t"`
 	LoadAvg      [3]float64                    `json:"la"`
 	Battery      [2]uint8                      `json:"bat"`
+	ExtraFs      map[string]SystemAlertFsStats `json:"efs"`
 }
 
 type SystemAlertGPUData struct {
@@ -58,7 +64,7 @@ type SystemAlertGPUData struct {
 
 type SystemAlertData struct {
 	systemRecord *core.Record
-	alertRecord  *core.Record
+	alertData    CachedAlertData
 	name         string
 	unit         string
 	val          float64
@@ -92,12 +98,10 @@ var supportsTitle = map[string]struct{}{
 // NewAlertManager creates a new AlertManager instance.
 func NewAlertManager(app hubLike) *AlertManager {
 	am := &AlertManager{
-		hub:        app,
-		alertQueue: make(chan alertTask, 5),
-		stopChan:   make(chan struct{}),
+		hub:         app,
+		alertsCache: NewAlertsCache(app),
 	}
 	am.bindEvents()
-	go am.startWorker()
 	return am
 }
 
@@ -106,6 +110,19 @@ func (am *AlertManager) bindEvents() {
 	am.hub.OnRecordAfterUpdateSuccess("alerts").BindFunc(updateHistoryOnAlertUpdate)
 	am.hub.OnRecordAfterDeleteSuccess("alerts").BindFunc(resolveHistoryOnAlertDelete)
 	am.hub.OnRecordAfterUpdateSuccess("smart_devices").BindFunc(am.handleSmartDeviceAlert)
+
+	am.hub.OnServe().BindFunc(func(e *core.ServeEvent) error {
+		// Populate all alerts into cache on startup
+		_ = am.alertsCache.PopulateFromDB(true)
+
+		if err := resolveStatusAlerts(e.App); err != nil {
+			e.App.Logger().Error("Failed to resolve stale status alerts", "err", err)
+		}
+		if err := am.restorePendingStatusAlerts(); err != nil {
+			e.App.Logger().Error("Failed to restore pending status alerts", "err", err)
+		}
+		return e.Next()
+	})
 }
 
 // IsNotificationSilenced checks if a notification should be silenced based on configured quiet hours
@@ -259,13 +276,14 @@ func (am *AlertManager) SendShoutrrrAlert(notificationUrl, title, message, link,
 	}
 
 	// Add link
-	if scheme == "ntfy" {
+	switch scheme {
+	case "ntfy":
 		queryParams.Add("Actions", fmt.Sprintf("view, %s, %s", linkText, link))
-	} else if scheme == "lark" {
+	case "lark":
 		queryParams.Add("link", link)
-	} else if scheme == "bark" {
+	case "bark":
 		queryParams.Add("url", link)
-	} else {
+	default:
 		message += "\n\n" + link
 	}
 
@@ -284,17 +302,12 @@ func (am *AlertManager) SendShoutrrrAlert(notificationUrl, title, message, link,
 	return nil
 }
 
-func (am *AlertManager) SendTestNotification(e *core.RequestEvent) error {
-	var data struct {
-		URL string `json:"url"`
-	}
-	err := e.BindBody(&data)
-	if err != nil || data.URL == "" {
-		return e.BadRequestError("URL is required", err)
-	}
-	err = am.SendShoutrrrAlert(data.URL, "Test Alert", "This is a notification from Beszel.", am.hub.Settings().Meta.AppURL, "View Beszel")
+// setAlertTriggered updates the "triggered" status of an alert record in the database
+func (am *AlertManager) setAlertTriggered(alert CachedAlertData, triggered bool) error {
+	alertRecord, err := am.hub.FindRecordById("alerts", alert.Id)
 	if err != nil {
-		return e.JSON(200, map[string]string{"err": err.Error()})
+		return err
 	}
-	return e.JSON(200, map[string]bool{"err": false})
+	alertRecord.Set("triggered", triggered)
+	return am.hub.Save(alertRecord)
 }

internal/alerts/alerts_api.go

@@ -3,7 +3,11 @@ package alerts
 import (
 	"database/sql"
 	"errors"
+	"net"
 	"net/http"
+	"net/url"
+	"slices"
+	"strings"
 
 	"github.com/pocketbase/dbx"
 	"github.com/pocketbase/pocketbase/core"
@@ -117,3 +121,72 @@ func DeleteUserAlerts(e *core.RequestEvent) error {
 
 	return e.JSON(http.StatusOK, map[string]any{"success": true, "count": numDeleted})
 }
+
+// SendTestNotification handles API request to send a test notification to a specified Shoutrrr URL
+func (am *AlertManager) SendTestNotification(e *core.RequestEvent) error {
+	var data struct {
+		URL string `json:"url"`
+	}
+	err := e.BindBody(&data)
+	if err != nil || data.URL == "" {
+		return e.BadRequestError("URL is required", err)
+	}
+	// Only allow admins to send test notifications to internal URLs
+	if !e.Auth.IsSuperuser() && e.Auth.GetString("role") != "admin" {
+		internalURL, err := isInternalURL(data.URL)
+		if err != nil {
+			return e.BadRequestError(err.Error(), nil)
+		}
+		if internalURL {
+			return e.ForbiddenError("Only admins can send to internal destinations", nil)
+		}
+	}
+	err = am.SendShoutrrrAlert(data.URL, "Test Alert", "This is a notification from Beszel.", am.hub.Settings().Meta.AppURL, "View Beszel")
+	if err != nil {
+		return e.JSON(200, map[string]string{"err": err.Error()})
+	}
+	return e.JSON(200, map[string]bool{"err": false})
+}
+
+// isInternalURL checks if the given shoutrrr URL points to an internal destination (localhost or private IP)
+func isInternalURL(rawURL string) (bool, error) {
+	parsedURL, err := url.Parse(rawURL)
+	if err != nil {
+		return false, err
+	}
+
+	host := parsedURL.Hostname()
+	if host == "" {
+		return false, nil
+	}
+
+	if strings.EqualFold(host, "localhost") {
+		return true, nil
+	}
+
+	if ip := net.ParseIP(host); ip != nil {
+		return isInternalIP(ip), nil
+	}
+
+	// Some Shoutrrr URLs use the host position for service identifiers rather than a
+	// network hostname (for example, discord://token@webhookid). Restrict DNS lookups
+	// to names that look like actual hostnames so valid service URLs keep working.
+	if !strings.Contains(host, ".") {
+		return false, nil
+	}
+
+	ips, err := net.LookupIP(host)
+	if err != nil {
+		return false, nil
+	}
+
+	if slices.ContainsFunc(ips, isInternalIP) {
+		return true, nil
+	}
+
+	return false, nil
+}
+
+func isInternalIP(ip net.IP) bool {
+	return ip.IsPrivate() || ip.IsLoopback() || ip.IsUnspecified()
+}

internal/alerts/alerts_api_test.go

@@ -0,0 +1,501 @@
+//go:build testing
+
+package alerts_test
+
+import (
+	"bytes"
+	"encoding/json"
+	"io"
+	"net/http"
+	"strings"
+	"testing"
+
+	"github.com/henrygd/beszel/internal/alerts"
+	beszelTests "github.com/henrygd/beszel/internal/tests"
+	pbTests "github.com/pocketbase/pocketbase/tests"
+
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/core"
+
+	"github.com/stretchr/testify/assert"
+)
+
+// marshal to json and return an io.Reader (for use in ApiScenario.Body)
+func jsonReader(v any) io.Reader {
+	data, err := json.Marshal(v)
+	if err != nil {
+		panic(err)
+	}
+	return bytes.NewReader(data)
+}
+
+func TestIsInternalURL(t *testing.T) {
+	testCases := []struct {
+		name     string
+		url      string
+		internal bool
+	}{
+		{name: "loopback ipv4", url: "generic://127.0.0.1", internal: true},
+		{name: "localhost hostname", url: "generic://localhost", internal: true},
+		{name: "localhost hostname", url: "generic+http://localhost/api/v1/postStuff", internal: true},
+		{name: "localhost hostname", url: "generic+http://127.0.0.1:8080/api/v1/postStuff", internal: true},
+		{name: "localhost hostname", url: "generic+https://beszel.dev/api/v1/postStuff", internal: false},
+		{name: "public ipv4", url: "generic://8.8.8.8", internal: false},
+		{name: "token style service url", url: "discord://abc123@123456789", internal: false},
+		{name: "single label service url", url: "slack://token@team/channel", internal: false},
+	}
+
+	for _, testCase := range testCases {
+		t.Run(testCase.name, func(t *testing.T) {
+			internal, err := alerts.IsInternalURL(testCase.url)
+			assert.NoError(t, err)
+			assert.Equal(t, testCase.internal, internal)
+		})
+	}
+}
+
+func TestUserAlertsApi(t *testing.T) {
+	hub, _ := beszelTests.NewTestHub(t.TempDir())
+	defer hub.Cleanup()
+
+	hub.StartHub()
+
+	user1, _ := beszelTests.CreateUser(hub, "alertstest@example.com", "password")
+	user1Token, _ := user1.NewAuthToken()
+
+	user2, _ := beszelTests.CreateUser(hub, "alertstest2@example.com", "password")
+	user2Token, _ := user2.NewAuthToken()
+
+	system1, _ := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "system1",
+		"users": []string{user1.Id},
+		"host":  "127.0.0.1",
+	})
+
+	system2, _ := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "system2",
+		"users": []string{user1.Id, user2.Id},
+		"host":  "127.0.0.2",
+	})
+
+	userRecords, _ := hub.CountRecords("users")
+	assert.EqualValues(t, 2, userRecords, "all users should be created")
+
+	systemRecords, _ := hub.CountRecords("systems")
+	assert.EqualValues(t, 2, systemRecords, "all systems should be created")
+
+	testAppFactory := func(t testing.TB) *pbTests.TestApp {
+		return hub.TestApp
+	}
+
+	scenarios := []beszelTests.ApiScenario{
+		// {
+		// 	Name:            "GET not implemented - returns index",
+		// 	Method:          http.MethodGet,
+		// 	URL:             "/api/beszel/user-alerts",
+		// 	ExpectedStatus:  200,
+		// 	ExpectedContent: []string{"<html ", "globalThis.BESZEL"},
+		// 	TestAppFactory:  testAppFactory,
+		// },
+		{
+			Name:            "POST no auth",
+			Method:          http.MethodPost,
+			URL:             "/api/beszel/user-alerts",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "POST no body",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  400,
+			ExpectedContent: []string{"Bad data"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "POST bad data",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  400,
+			ExpectedContent: []string{"Bad data"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"invalidField": "this should cause validation error",
+				"threshold":    "not a number",
+			}),
+		},
+		{
+			Name:   "POST malformed JSON",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  400,
+			ExpectedContent: []string{"Bad data"},
+			TestAppFactory:  testAppFactory,
+			Body:            strings.NewReader(`{"alertType": "cpu", "threshold": 80, "enabled": true,}`),
+		},
+		{
+			Name:   "POST valid alert data multiple systems",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":      "CPU",
+				"value":     69,
+				"min":       9,
+				"systems":   []string{system1.Id, system2.Id},
+				"overwrite": false,
+			}),
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				// check total alerts
+				alerts, _ := app.CountRecords("alerts")
+				assert.EqualValues(t, 2, alerts, "should have 2 alerts")
+				// check alert has correct values
+				matchingAlerts, _ := app.CountRecords("alerts", dbx.HashExp{"name": "CPU", "user": user1.Id, "system": system1.Id, "value": 69, "min": 9})
+				assert.EqualValues(t, 1, matchingAlerts, "should have 1 alert")
+			},
+		},
+		{
+			Name:   "POST valid alert data single system",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "Memory",
+				"systems": []string{system1.Id},
+				"value":   90,
+				"min":     10,
+			}),
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				user1Alerts, _ := app.CountRecords("alerts", dbx.HashExp{"user": user1.Id})
+				assert.EqualValues(t, 3, user1Alerts, "should have 3 alerts")
+			},
+		},
+		{
+			Name:   "Overwrite: false, should not overwrite existing alert",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":      "CPU",
+				"value":     45,
+				"min":       5,
+				"systems":   []string{system1.Id},
+				"overwrite": false,
+			}),
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.ClearCollection(t, app, "alerts")
+				beszelTests.CreateRecord(app, "alerts", map[string]any{
+					"name":   "CPU",
+					"system": system1.Id,
+					"user":   user1.Id,
+					"value":  80,
+					"min":    10,
+				})
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				alerts, _ := app.CountRecords("alerts")
+				assert.EqualValues(t, 1, alerts, "should have 1 alert")
+				alert, _ := app.FindFirstRecordByFilter("alerts", "name = 'CPU' && user = {:user}", dbx.Params{"user": user1.Id})
+				assert.EqualValues(t, 80, alert.Get("value"), "should have 80 as value")
+			},
+		},
+		{
+			Name:   "Overwrite: true, should overwrite existing alert",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user2Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":      "CPU",
+				"value":     45,
+				"min":       5,
+				"systems":   []string{system2.Id},
+				"overwrite": true,
+			}),
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.ClearCollection(t, app, "alerts")
+				beszelTests.CreateRecord(app, "alerts", map[string]any{
+					"name":   "CPU",
+					"system": system2.Id,
+					"user":   user2.Id,
+					"value":  80,
+					"min":    10,
+				})
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				alerts, _ := app.CountRecords("alerts")
+				assert.EqualValues(t, 1, alerts, "should have 1 alert")
+				alert, _ := app.FindFirstRecordByFilter("alerts", "name = 'CPU' && user = {:user}", dbx.Params{"user": user2.Id})
+				assert.EqualValues(t, 45, alert.Get("value"), "should have 45 as value")
+			},
+		},
+		{
+			Name:            "DELETE no auth",
+			Method:          http.MethodDelete,
+			URL:             "/api/beszel/user-alerts",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "CPU",
+				"systems": []string{system1.Id},
+			}),
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.ClearCollection(t, app, "alerts")
+				beszelTests.CreateRecord(app, "alerts", map[string]any{
+					"name":   "CPU",
+					"system": system1.Id,
+					"user":   user1.Id,
+					"value":  80,
+					"min":    10,
+				})
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				alerts, _ := app.CountRecords("alerts")
+				assert.EqualValues(t, 1, alerts, "should have 1 alert")
+			},
+		},
+		{
+			Name:   "DELETE alert",
+			Method: http.MethodDelete,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"count\":1", "\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "CPU",
+				"systems": []string{system1.Id},
+			}),
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.ClearCollection(t, app, "alerts")
+				beszelTests.CreateRecord(app, "alerts", map[string]any{
+					"name":   "CPU",
+					"system": system1.Id,
+					"user":   user1.Id,
+					"value":  80,
+					"min":    10,
+				})
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				alerts, _ := app.CountRecords("alerts")
+				assert.Zero(t, alerts, "should have 0 alerts")
+			},
+		},
+		{
+			Name:   "DELETE alert multiple systems",
+			Method: http.MethodDelete,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user1Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"count\":2", "\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "Memory",
+				"systems": []string{system1.Id, system2.Id},
+			}),
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.ClearCollection(t, app, "alerts")
+				for _, systemId := range []string{system1.Id, system2.Id} {
+					_, err := beszelTests.CreateRecord(app, "alerts", map[string]any{
+						"name":   "Memory",
+						"system": systemId,
+						"user":   user1.Id,
+						"value":  90,
+						"min":    10,
+					})
+					assert.NoError(t, err, "should create alert")
+				}
+				alerts, _ := app.CountRecords("alerts")
+				assert.EqualValues(t, 2, alerts, "should have 2 alerts")
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				alerts, _ := app.CountRecords("alerts")
+				assert.Zero(t, alerts, "should have 0 alerts")
+			},
+		},
+		{
+			Name:   "User 2 should not be able to delete alert of user 1",
+			Method: http.MethodDelete,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": user2Token,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"count\":1", "\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "CPU",
+				"systems": []string{system2.Id},
+			}),
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.ClearCollection(t, app, "alerts")
+				for _, user := range []string{user1.Id, user2.Id} {
+					beszelTests.CreateRecord(app, "alerts", map[string]any{
+						"name":   "CPU",
+						"system": system2.Id,
+						"user":   user,
+						"value":  80,
+						"min":    10,
+					})
+				}
+				alerts, _ := app.CountRecords("alerts")
+				assert.EqualValues(t, 2, alerts, "should have 2 alerts")
+				user1AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user1.Id})
+				assert.EqualValues(t, 1, user1AlertCount, "should have 1 alert")
+				user2AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user2.Id})
+				assert.EqualValues(t, 1, user2AlertCount, "should have 1 alert")
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				user1AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user1.Id})
+				assert.EqualValues(t, 1, user1AlertCount, "should have 1 alert")
+				user2AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user2.Id})
+				assert.Zero(t, user2AlertCount, "should have 0 alerts")
+			},
+		},
+	}
+
+	for _, scenario := range scenarios {
+		scenario.Test(t)
+	}
+}
+func TestSendTestNotification(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	userToken, err := user.NewAuthToken()
+
+	adminUser, err := beszelTests.CreateUserWithRole(hub, "admin@example.com", "password123", "admin")
+	assert.NoError(t, err, "Failed to create admin user")
+	adminUserToken, err := adminUser.NewAuthToken()
+
+	superuser, err := beszelTests.CreateSuperuser(hub, "superuser@example.com", "password123")
+	assert.NoError(t, err, "Failed to create superuser")
+	superuserToken, err := superuser.NewAuthToken()
+	assert.NoError(t, err, "Failed to create superuser auth token")
+
+	testAppFactory := func(t testing.TB) *pbTests.TestApp {
+		return hub.TestApp
+	}
+
+	scenarios := []beszelTests.ApiScenario{
+		{
+			Name:            "POST /test-notification - no auth should fail",
+			Method:          http.MethodPost,
+			URL:             "/api/beszel/test-notification",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"url": "generic://127.0.0.1",
+			}),
+		},
+		{
+			Name:           "POST /test-notification - with external auth should succeed",
+			Method:         http.MethodPost,
+			URL:            "/api/beszel/test-notification",
+			TestAppFactory: testAppFactory,
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			Body: jsonReader(map[string]any{
+				"url": "generic://8.8.8.8",
+			}),
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"err\":"},
+		},
+		{
+			Name:           "POST /test-notification - local url with user auth should fail",
+			Method:         http.MethodPost,
+			URL:            "/api/beszel/test-notification",
+			TestAppFactory: testAppFactory,
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			Body: jsonReader(map[string]any{
+				"url": "generic://localhost:8010",
+			}),
+			ExpectedStatus:  403,
+			ExpectedContent: []string{"Only admins"},
+		},
+		{
+			Name:           "POST /test-notification - internal url with user auth should fail",
+			Method:         http.MethodPost,
+			URL:            "/api/beszel/test-notification",
+			TestAppFactory: testAppFactory,
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			Body: jsonReader(map[string]any{
+				"url": "generic+http://192.168.0.5",
+			}),
+			ExpectedStatus:  403,
+			ExpectedContent: []string{"Only admins"},
+		},
+		{
+			Name:           "POST /test-notification - internal url with admin auth should succeed",
+			Method:         http.MethodPost,
+			URL:            "/api/beszel/test-notification",
+			TestAppFactory: testAppFactory,
+			Headers: map[string]string{
+				"Authorization": adminUserToken,
+			},
+			Body: jsonReader(map[string]any{
+				"url": "generic://127.0.0.1",
+			}),
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"err\":"},
+		},
+		{
+			Name:           "POST /test-notification - internal url with superuser auth should succeed",
+			Method:         http.MethodPost,
+			URL:            "/api/beszel/test-notification",
+			TestAppFactory: testAppFactory,
+			Headers: map[string]string{
+				"Authorization": superuserToken,
+			},
+			Body: jsonReader(map[string]any{
+				"url": "generic://127.0.0.1",
+			}),
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"err\":"},
+		},
+	}
+
+	for _, scenario := range scenarios {
+		scenario.Test(t)
+	}
+}

internal/alerts/alerts_battery_test.go

@@ -1,5 +1,4 @@
 //go:build testing
-// +build testing
 
 package alerts_test
 

internal/alerts/alerts_cache.go

@@ -0,0 +1,177 @@
+package alerts
+
+import (
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/core"
+	"github.com/pocketbase/pocketbase/tools/store"
+)
+
+// CachedAlertData represents the relevant fields of an alert record for status checking and updates.
+type CachedAlertData struct {
+	Id        string
+	SystemID  string
+	UserID    string
+	Name      string
+	Value     float64
+	Triggered bool
+	Min       uint8
+	// Created   types.DateTime
+}
+
+func (a *CachedAlertData) PopulateFromRecord(record *core.Record) {
+	a.Id = record.Id
+	a.SystemID = record.GetString("system")
+	a.UserID = record.GetString("user")
+	a.Name = record.GetString("name")
+	a.Value = record.GetFloat("value")
+	a.Triggered = record.GetBool("triggered")
+	a.Min = uint8(record.GetInt("min"))
+	// a.Created = record.GetDateTime("created")
+}
+
+// AlertsCache provides an in-memory cache for system alerts.
+type AlertsCache struct {
+	app       core.App
+	store     *store.Store[string, *store.Store[string, CachedAlertData]]
+	populated bool
+}
+
+// NewAlertsCache creates a new instance of SystemAlertsCache.
+func NewAlertsCache(app core.App) *AlertsCache {
+	c := AlertsCache{
+		app:   app,
+		store: store.New(map[string]*store.Store[string, CachedAlertData]{}),
+	}
+	return c.bindEvents()
+}
+
+// bindEvents sets up event listeners to keep the cache in sync with database changes.
+func (c *AlertsCache) bindEvents() *AlertsCache {
+	c.app.OnRecordAfterUpdateSuccess("alerts").BindFunc(func(e *core.RecordEvent) error {
+		// c.Delete(e.Record.Original()) // this would be needed if the system field on an existing alert was changed, however we don't currently allow that in the UI so we'll leave it commented out
+		c.Update(e.Record)
+		return e.Next()
+	})
+	c.app.OnRecordAfterDeleteSuccess("alerts").BindFunc(func(e *core.RecordEvent) error {
+		c.Delete(e.Record)
+		return e.Next()
+	})
+	c.app.OnRecordAfterCreateSuccess("alerts").BindFunc(func(e *core.RecordEvent) error {
+		c.Update(e.Record)
+		return e.Next()
+	})
+	return c
+}
+
+// PopulateFromDB clears current entries and loads all alerts from the database into the cache.
+func (c *AlertsCache) PopulateFromDB(force bool) error {
+	if !force && c.populated {
+		return nil
+	}
+	records, err := c.app.FindAllRecords("alerts")
+	if err != nil {
+		return err
+	}
+	c.store.RemoveAll()
+	for _, record := range records {
+		c.Update(record)
+	}
+	c.populated = true
+	return nil
+}
+
+// Update adds or updates an alert record in the cache.
+func (c *AlertsCache) Update(record *core.Record) {
+	systemID := record.GetString("system")
+	if systemID == "" {
+		return
+	}
+	systemStore, ok := c.store.GetOk(systemID)
+	if !ok {
+		systemStore = store.New(map[string]CachedAlertData{})
+		c.store.Set(systemID, systemStore)
+	}
+	var ca CachedAlertData
+	ca.PopulateFromRecord(record)
+	systemStore.Set(record.Id, ca)
+}
+
+// Delete removes an alert record from the cache.
+func (c *AlertsCache) Delete(record *core.Record) {
+	systemID := record.GetString("system")
+	if systemID == "" {
+		return
+	}
+	if systemStore, ok := c.store.GetOk(systemID); ok {
+		systemStore.Remove(record.Id)
+	}
+}
+
+// GetSystemAlerts returns all alerts for the specified system, lazy-loading if necessary.
+func (c *AlertsCache) GetSystemAlerts(systemID string) []CachedAlertData {
+	systemStore, ok := c.store.GetOk(systemID)
+	if !ok {
+		// Populate cache for this system
+		records, err := c.app.FindAllRecords("alerts", dbx.NewExp("system={:system}", dbx.Params{"system": systemID}))
+		if err != nil {
+			return nil
+		}
+		systemStore = store.New(map[string]CachedAlertData{})
+		for _, record := range records {
+			var ca CachedAlertData
+			ca.PopulateFromRecord(record)
+			systemStore.Set(record.Id, ca)
+		}
+		c.store.Set(systemID, systemStore)
+	}
+	all := systemStore.GetAll()
+	alerts := make([]CachedAlertData, 0, len(all))
+	for _, alert := range all {
+		alerts = append(alerts, alert)
+	}
+	return alerts
+}
+
+// GetAlert returns a specific alert by its ID from the cache.
+func (c *AlertsCache) GetAlert(systemID, alertID string) (CachedAlertData, bool) {
+	if systemStore, ok := c.store.GetOk(systemID); ok {
+		return systemStore.GetOk(alertID)
+	}
+	return CachedAlertData{}, false
+}
+
+// GetAlertsByName returns all alerts of a specific type for the specified system.
+func (c *AlertsCache) GetAlertsByName(systemID, alertName string) []CachedAlertData {
+	allAlerts := c.GetSystemAlerts(systemID)
+	var alerts []CachedAlertData
+	for _, record := range allAlerts {
+		if record.Name == alertName {
+			alerts = append(alerts, record)
+		}
+	}
+	return alerts
+}
+
+// GetAlertsExcludingNames returns all alerts for the specified system excluding the given types.
+func (c *AlertsCache) GetAlertsExcludingNames(systemID string, excludedNames ...string) []CachedAlertData {
+	excludeMap := make(map[string]struct{})
+	for _, name := range excludedNames {
+		excludeMap[name] = struct{}{}
+	}
+	allAlerts := c.GetSystemAlerts(systemID)
+	var alerts []CachedAlertData
+	for _, record := range allAlerts {
+		if _, excluded := excludeMap[record.Name]; !excluded {
+			alerts = append(alerts, record)
+		}
+	}
+	return alerts
+}
+
+// Refresh returns the latest cached copy for an alert snapshot if it still exists.
+func (c *AlertsCache) Refresh(alert CachedAlertData) (CachedAlertData, bool) {
+	if alert.Id == "" {
+		return CachedAlertData{}, false
+	}
+	return c.GetAlert(alert.SystemID, alert.Id)
+}

internal/alerts/alerts_cache_test.go

@@ -0,0 +1,215 @@
+//go:build testing
+
+package alerts_test
+
+import (
+	"testing"
+
+	"github.com/henrygd/beszel/internal/alerts"
+	beszelTests "github.com/henrygd/beszel/internal/tests"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+func TestSystemAlertsCachePopulateAndFilter(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	systems, err := beszelTests.CreateSystems(hub, 2, user.Id, "up")
+	require.NoError(t, err)
+	system1 := systems[0]
+	system2 := systems[1]
+
+	statusAlert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":   "Status",
+		"system": system1.Id,
+		"user":   user.Id,
+		"min":    1,
+	})
+	require.NoError(t, err)
+
+	cpuAlert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":   "CPU",
+		"system": system1.Id,
+		"user":   user.Id,
+		"value":  80,
+		"min":    1,
+	})
+	require.NoError(t, err)
+
+	memoryAlert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":   "Memory",
+		"system": system2.Id,
+		"user":   user.Id,
+		"value":  90,
+		"min":    1,
+	})
+	require.NoError(t, err)
+
+	cache := alerts.NewAlertsCache(hub)
+	cache.PopulateFromDB(false)
+
+	statusAlerts := cache.GetAlertsByName(system1.Id, "Status")
+	require.Len(t, statusAlerts, 1)
+	assert.Equal(t, statusAlert.Id, statusAlerts[0].Id)
+
+	nonStatusAlerts := cache.GetAlertsExcludingNames(system1.Id, "Status")
+	require.Len(t, nonStatusAlerts, 1)
+	assert.Equal(t, cpuAlert.Id, nonStatusAlerts[0].Id)
+
+	system2Alerts := cache.GetSystemAlerts(system2.Id)
+	require.Len(t, system2Alerts, 1)
+	assert.Equal(t, memoryAlert.Id, system2Alerts[0].Id)
+}
+
+func TestSystemAlertsCacheLazyLoadUpdateAndDelete(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	systems, err := beszelTests.CreateSystems(hub, 1, user.Id, "up")
+	require.NoError(t, err)
+	systemRecord := systems[0]
+
+	statusAlert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":   "Status",
+		"system": systemRecord.Id,
+		"user":   user.Id,
+		"min":    1,
+	})
+	require.NoError(t, err)
+
+	cache := alerts.NewAlertsCache(hub)
+	require.Len(t, cache.GetSystemAlerts(systemRecord.Id), 1, "first lookup should lazy-load alerts for the system")
+
+	cpuAlert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":   "CPU",
+		"system": systemRecord.Id,
+		"user":   user.Id,
+		"value":  80,
+		"min":    1,
+	})
+	require.NoError(t, err)
+
+	cache.Update(cpuAlert)
+
+	nonStatusAlerts := cache.GetAlertsExcludingNames(systemRecord.Id, "Status")
+	require.Len(t, nonStatusAlerts, 1)
+	assert.Equal(t, cpuAlert.Id, nonStatusAlerts[0].Id)
+
+	cache.Delete(statusAlert)
+	assert.Empty(t, cache.GetAlertsByName(systemRecord.Id, "Status"), "deleted alerts should be removed from the in-memory cache")
+}
+
+func TestSystemAlertsCacheRefreshReturnsLatestCopy(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	systems, err := beszelTests.CreateSystems(hub, 1, user.Id, "up")
+	require.NoError(t, err)
+	system := systems[0]
+
+	alert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":      "Status",
+		"system":    system.Id,
+		"user":      user.Id,
+		"min":       1,
+		"triggered": false,
+	})
+	require.NoError(t, err)
+
+	cache := alerts.NewAlertsCache(hub)
+	snapshot := cache.GetSystemAlerts(system.Id)[0]
+	assert.False(t, snapshot.Triggered)
+
+	alert.Set("triggered", true)
+	require.NoError(t, hub.Save(alert))
+
+	refreshed, ok := cache.Refresh(snapshot)
+	require.True(t, ok)
+	assert.Equal(t, snapshot.Id, refreshed.Id)
+	assert.True(t, refreshed.Triggered, "refresh should return the updated cached value rather than the stale snapshot")
+
+	require.NoError(t, hub.Delete(alert))
+	_, ok = cache.Refresh(snapshot)
+	assert.False(t, ok, "refresh should report false when the cached alert no longer exists")
+}
+
+func TestAlertManagerCacheLifecycle(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	systems, err := beszelTests.CreateSystems(hub, 1, user.Id, "up")
+	require.NoError(t, err)
+	system := systems[0]
+
+	// Create an alert
+	alert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":   "CPU",
+		"system": system.Id,
+		"user":   user.Id,
+		"value":  80,
+		"min":    1,
+	})
+	require.NoError(t, err)
+
+	am := hub.AlertManager
+	cache := am.GetSystemAlertsCache()
+
+	// Verify it's in cache (it should be since CreateRecord triggers the event)
+	assert.Len(t, cache.GetSystemAlerts(system.Id), 1)
+	assert.Equal(t, alert.Id, cache.GetSystemAlerts(system.Id)[0].Id)
+	assert.EqualValues(t, 80, cache.GetSystemAlerts(system.Id)[0].Value)
+
+	// Update the alert through PocketBase to trigger events
+	alert.Set("value", 85)
+	require.NoError(t, hub.Save(alert))
+
+	// Check if updated value is reflected (or at least that it's still there)
+	cachedAlerts := cache.GetSystemAlerts(system.Id)
+	assert.Len(t, cachedAlerts, 1)
+	assert.EqualValues(t, 85, cachedAlerts[0].Value)
+
+	// Delete the alert through PocketBase to trigger events
+	require.NoError(t, hub.Delete(alert))
+
+	// Verify it's removed from cache
+	assert.Empty(t, cache.GetSystemAlerts(system.Id), "alert should be removed from cache after PocketBase delete")
+}
+
+// func TestAlertManagerCacheMovesAlertToNewSystemOnUpdate(t *testing.T) {
+// 	hub, user := beszelTests.GetHubWithUser(t)
+// 	defer hub.Cleanup()
+
+// 	systems, err := beszelTests.CreateSystems(hub, 2, user.Id, "up")
+// 	require.NoError(t, err)
+// 	system1 := systems[0]
+// 	system2 := systems[1]
+
+// 	alert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+// 		"name":   "CPU",
+// 		"system": system1.Id,
+// 		"user":   user.Id,
+// 		"value":  80,
+// 		"min":    1,
+// 	})
+// 	require.NoError(t, err)
+
+// 	am := hub.AlertManager
+// 	cache := am.GetSystemAlertsCache()
+
+// 	// Initially in system1 cache
+// 	assert.Len(t, cache.Get(system1.Id), 1)
+// 	assert.Empty(t, cache.Get(system2.Id))
+
+// 	// Move alert to system2
+// 	alert.Set("system", system2.Id)
+// 	require.NoError(t, hub.Save(alert))
+
+// 	// DEBUG: print if it is found
+// 	// fmt.Printf("system1 alerts after update: %v\n", cache.Get(system1.Id))
+
+// 	// Should be removed from system1 and present in system2
+// 	assert.Empty(t, cache.GetType(system1.Id, "CPU"), "updated alerts should be evicted from the previous system cache")
+// 	require.Len(t, cache.Get(system2.Id), 1)
+// 	assert.Equal(t, alert.Id, cache.Get(system2.Id)[0].Id)
+// }

internal/alerts/alerts_disk_test.go

@@ -0,0 +1,155 @@
+//go:build testing
+
+package alerts_test
+
+import (
+	"encoding/json"
+	"testing"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/system"
+	beszelTests "github.com/henrygd/beszel/internal/tests"
+
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/tools/types"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// TestDiskAlertExtraFsMultiMinute tests that multi-minute disk alerts correctly use
+// historical per-minute values for extra (non-root) filesystems, not the current live snapshot.
+func TestDiskAlertExtraFsMultiMinute(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	systems, err := beszelTests.CreateSystems(hub, 1, user.Id, "up")
+	require.NoError(t, err)
+	systemRecord := systems[0]
+
+	// Disk alert: threshold 80%, min=2 (requires historical averaging)
+	diskAlert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":   "Disk",
+		"system": systemRecord.Id,
+		"user":   user.Id,
+		"value":  80, // threshold: 80%
+		"min":    2,  // 2 minutes - requires historical averaging
+	})
+	require.NoError(t, err)
+	assert.False(t, diskAlert.GetBool("triggered"), "Alert should not be triggered initially")
+
+	am := hub.GetAlertManager()
+	now := time.Now().UTC()
+
+	extraFsHigh := map[string]*system.FsStats{
+		"/mnt/data": {DiskTotal: 1000, DiskUsed: 920}, // 92% - above threshold
+	}
+
+	// Insert 4 historical records spread over 3 minutes (same pattern as battery tests).
+	// The oldest record must predate (now - 2min) so the alert time window is valid.
+	recordTimes := []time.Duration{
+		-180 * time.Second, // 3 min ago - anchors oldest record before alert.time
+		-90 * time.Second,
+		-60 * time.Second,
+		-30 * time.Second,
+	}
+
+	for _, offset := range recordTimes {
+		stats := system.Stats{
+			DiskPct: 30, // root disk at 30% - below threshold
+			ExtraFs: extraFsHigh,
+		}
+		statsJSON, _ := json.Marshal(stats)
+
+		recordTime := now.Add(offset)
+		record, err := beszelTests.CreateRecord(hub, "system_stats", map[string]any{
+			"system": systemRecord.Id,
+			"type":   "1m",
+			"stats":  string(statsJSON),
+		})
+		require.NoError(t, err)
+		record.SetRaw("created", recordTime.Format(types.DefaultDateLayout))
+		err = hub.SaveNoValidate(record)
+		require.NoError(t, err)
+	}
+
+	combinedDataHigh := &system.CombinedData{
+		Stats: system.Stats{
+			DiskPct: 30,
+			ExtraFs: extraFsHigh,
+		},
+		Info: system.Info{
+			DiskPct: 30,
+		},
+	}
+
+	systemRecord.Set("updated", now)
+	err = hub.SaveNoValidate(systemRecord)
+	require.NoError(t, err)
+
+	err = am.HandleSystemAlerts(systemRecord, combinedDataHigh)
+	require.NoError(t, err)
+
+	time.Sleep(20 * time.Millisecond)
+
+	diskAlert, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": diskAlert.Id})
+	require.NoError(t, err)
+	assert.True(t, diskAlert.GetBool("triggered"),
+		"Alert SHOULD be triggered when extra disk average (92%%) exceeds threshold (80%%)")
+
+	// --- Resolution: extra disk drops to 50%, alert should resolve ---
+
+	extraFsLow := map[string]*system.FsStats{
+		"/mnt/data": {DiskTotal: 1000, DiskUsed: 500}, // 50% - below threshold
+	}
+
+	newNow := now.Add(2 * time.Minute)
+	recordTimesLow := []time.Duration{
+		-180 * time.Second,
+		-90 * time.Second,
+		-60 * time.Second,
+		-30 * time.Second,
+	}
+
+	for _, offset := range recordTimesLow {
+		stats := system.Stats{
+			DiskPct: 30,
+			ExtraFs: extraFsLow,
+		}
+		statsJSON, _ := json.Marshal(stats)
+
+		recordTime := newNow.Add(offset)
+		record, err := beszelTests.CreateRecord(hub, "system_stats", map[string]any{
+			"system": systemRecord.Id,
+			"type":   "1m",
+			"stats":  string(statsJSON),
+		})
+		require.NoError(t, err)
+		record.SetRaw("created", recordTime.Format(types.DefaultDateLayout))
+		err = hub.SaveNoValidate(record)
+		require.NoError(t, err)
+	}
+
+	combinedDataLow := &system.CombinedData{
+		Stats: system.Stats{
+			DiskPct: 30,
+			ExtraFs: extraFsLow,
+		},
+		Info: system.Info{
+			DiskPct: 30,
+		},
+	}
+
+	systemRecord.Set("updated", newNow)
+	err = hub.SaveNoValidate(systemRecord)
+	require.NoError(t, err)
+
+	err = am.HandleSystemAlerts(systemRecord, combinedDataLow)
+	require.NoError(t, err)
+
+	time.Sleep(20 * time.Millisecond)
+
+	diskAlert, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": diskAlert.Id})
+	require.NoError(t, err)
+	assert.False(t, diskAlert.GetBool("triggered"),
+		"Alert should be resolved when extra disk average (50%%) drops below threshold (80%%)")
+}

internal/alerts/alerts_quiet_hours_test.go

@@ -1,5 +1,4 @@
 //go:build testing
-// +build testing
 
 package alerts_test
 
@@ -50,7 +49,7 @@ func TestAlertSilencedOneTime(t *testing.T) {
 
 	// Get alert manager
 	am := alerts.NewAlertManager(hub)
-	defer am.StopWorker()
+	defer am.Stop()
 
 	// Test that alert is silenced
 	silenced := am.IsNotificationSilenced(user.Id, system.Id)
@@ -107,7 +106,7 @@ func TestAlertSilencedDaily(t *testing.T) {
 
 	// Get alert manager
 	am := alerts.NewAlertManager(hub)
-	defer am.StopWorker()
+	defer am.Stop()
 
 	// Get current hour and create a window that includes current time
 	now := time.Now().UTC()
@@ -171,7 +170,7 @@ func TestAlertSilencedDailyMidnightCrossing(t *testing.T) {
 
 	// Get alert manager
 	am := alerts.NewAlertManager(hub)
-	defer am.StopWorker()
+	defer am.Stop()
 
 	// Create a window that crosses midnight: 22:00 - 02:00
 	startTime := time.Date(2000, 1, 1, 22, 0, 0, 0, time.UTC)
@@ -212,7 +211,7 @@ func TestAlertSilencedGlobal(t *testing.T) {
 
 	// Get alert manager
 	am := alerts.NewAlertManager(hub)
-	defer am.StopWorker()
+	defer am.Stop()
 
 	// Create a global quiet hours window (no system specified)
 	now := time.Now().UTC()
@@ -251,7 +250,7 @@ func TestAlertSilencedSystemSpecific(t *testing.T) {
 
 	// Get alert manager
 	am := alerts.NewAlertManager(hub)
-	defer am.StopWorker()
+	defer am.Stop()
 
 	// Create a system-specific quiet hours window for system1 only
 	now := time.Now().UTC()
@@ -297,7 +296,7 @@ func TestAlertSilencedMultiUser(t *testing.T) {
 
 	// Get alert manager
 	am := alerts.NewAlertManager(hub)
-	defer am.StopWorker()
+	defer am.Stop()
 
 	// Create a quiet hours window for user1 only
 	now := time.Now().UTC()
@@ -418,7 +417,7 @@ func TestAlertSilencedNoWindows(t *testing.T) {
 
 	// Get alert manager
 	am := alerts.NewAlertManager(hub)
-	defer am.StopWorker()
+	defer am.Stop()
 
 	// Without any quiet hours windows, alert should NOT be silenced
 	silenced := am.IsNotificationSilenced(user.Id, system.Id)

internal/alerts/alerts_smart.go

@@ -2,18 +2,18 @@ package alerts
 
 import (
 	"fmt"
+	"strings"
 
 	"github.com/pocketbase/pocketbase/core"
 )
 
-// handleSmartDeviceAlert sends alerts when a SMART device state changes from PASSED to FAILED.
+// handleSmartDeviceAlert sends alerts when a SMART device state worsens into WARNING/FAILED.
 // This is automatic and does not require user opt-in.
 func (am *AlertManager) handleSmartDeviceAlert(e *core.RecordEvent) error {
 	oldState := e.Record.Original().GetString("state")
 	newState := e.Record.GetString("state")
 
-	// Only alert when transitioning from PASSED to FAILED
-	if oldState != "PASSED" || newState != "FAILED" {
+	if !shouldSendSmartDeviceAlert(oldState, newState) {
 		return e.Next()
 	}
 
@@ -32,14 +32,15 @@ func (am *AlertManager) handleSmartDeviceAlert(e *core.RecordEvent) error {
 	systemName := systemRecord.GetString("name")
 	deviceName := e.Record.GetString("name")
 	model := e.Record.GetString("model")
+	statusLabel := smartStateLabel(newState)
 
 	// Build alert message
-	title := fmt.Sprintf("SMART failure on %s: %s \U0001F534", systemName, deviceName)
+	title := fmt.Sprintf("SMART %s on %s: %s %s", statusLabel, systemName, deviceName, smartStateEmoji(newState))
 	var message string
 	if model != "" {
-		message = fmt.Sprintf("Disk %s (%s) SMART status changed to FAILED", deviceName, model)
+		message = fmt.Sprintf("Disk %s (%s) SMART status changed to %s", deviceName, model, newState)
 	} else {
-		message = fmt.Sprintf("Disk %s SMART status changed to FAILED", deviceName)
+		message = fmt.Sprintf("Disk %s SMART status changed to %s", deviceName, newState)
 	}
 
 	// Get users associated with the system
@@ -65,3 +66,42 @@ func (am *AlertManager) handleSmartDeviceAlert(e *core.RecordEvent) error {
 	return e.Next()
 }
 
+func shouldSendSmartDeviceAlert(oldState, newState string) bool {
+	oldSeverity := smartStateSeverity(oldState)
+	newSeverity := smartStateSeverity(newState)
+
+	// Ignore unknown states and recoveries; only alert on worsening transitions
+	// from known-good/degraded states into WARNING/FAILED.
+	return oldSeverity >= 1 && newSeverity > oldSeverity
+}
+
+func smartStateSeverity(state string) int {
+	switch state {
+	case "PASSED":
+		return 1
+	case "WARNING":
+		return 2
+	case "FAILED":
+		return 3
+	default:
+		return 0
+	}
+}
+
+func smartStateEmoji(state string) string {
+	switch state {
+	case "WARNING":
+		return "\U0001F7E0"
+	default:
+		return "\U0001F534"
+	}
+}
+
+func smartStateLabel(state string) string {
+	switch state {
+	case "FAILED":
+		return "failure"
+	default:
+		return strings.ToLower(state)
+	}
+}

internal/alerts/alerts_smart_test.go

@@ -1,5 +1,4 @@
 //go:build testing
-// +build testing
 
 package alerts_test
 
@@ -58,6 +57,74 @@ func TestSmartDeviceAlert(t *testing.T) {
 	assert.Contains(t, lastMessage.Text, "FAILED")
 }
 
+func TestSmartDeviceAlertPassedToWarning(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	system, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "test-system",
+		"users": []string{user.Id},
+		"host":  "127.0.0.1",
+	})
+	assert.NoError(t, err)
+
+	smartDevice, err := beszelTests.CreateRecord(hub, "smart_devices", map[string]any{
+		"system": system.Id,
+		"name":   "/dev/mmcblk0",
+		"model":  "eMMC",
+		"state":  "PASSED",
+	})
+	assert.NoError(t, err)
+
+	smartDevice, err = hub.FindRecordById("smart_devices", smartDevice.Id)
+	assert.NoError(t, err)
+
+	smartDevice.Set("state", "WARNING")
+	err = hub.Save(smartDevice)
+	assert.NoError(t, err)
+
+	time.Sleep(50 * time.Millisecond)
+
+	assert.EqualValues(t, 1, hub.TestMailer.TotalSend(), "should have 1 email sent after state changed to WARNING")
+	lastMessage := hub.TestMailer.LastMessage()
+	assert.Contains(t, lastMessage.Subject, "SMART warning on test-system")
+	assert.Contains(t, lastMessage.Text, "WARNING")
+}
+
+func TestSmartDeviceAlertWarningToFailed(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	system, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "test-system",
+		"users": []string{user.Id},
+		"host":  "127.0.0.1",
+	})
+	assert.NoError(t, err)
+
+	smartDevice, err := beszelTests.CreateRecord(hub, "smart_devices", map[string]any{
+		"system": system.Id,
+		"name":   "/dev/mmcblk0",
+		"model":  "eMMC",
+		"state":  "WARNING",
+	})
+	assert.NoError(t, err)
+
+	smartDevice, err = hub.FindRecordById("smart_devices", smartDevice.Id)
+	assert.NoError(t, err)
+
+	smartDevice.Set("state", "FAILED")
+	err = hub.Save(smartDevice)
+	assert.NoError(t, err)
+
+	time.Sleep(50 * time.Millisecond)
+
+	assert.EqualValues(t, 1, hub.TestMailer.TotalSend(), "should have 1 email sent after state changed from WARNING to FAILED")
+	lastMessage := hub.TestMailer.LastMessage()
+	assert.Contains(t, lastMessage.Subject, "SMART failure on test-system")
+	assert.Contains(t, lastMessage.Text, "FAILED")
+}
+
 func TestSmartDeviceAlertNoAlertOnNonPassedToFailed(t *testing.T) {
 	hub, user := beszelTests.GetHubWithUser(t)
 	defer hub.Cleanup()
@@ -83,7 +150,8 @@ func TestSmartDeviceAlertNoAlertOnNonPassedToFailed(t *testing.T) {
 	smartDevice, err = hub.FindRecordById("smart_devices", smartDevice.Id)
 	assert.NoError(t, err)
 
-	// Update the state from UNKNOWN to FAILED - should NOT trigger alert
+	// Update the state from UNKNOWN to FAILED - should NOT trigger alert.
+	// We only alert from known healthy/degraded states.
 	smartDevice.Set("state", "FAILED")
 	err = hub.Save(smartDevice)
 	assert.NoError(t, err)

internal/alerts/alerts_status.go

@@ -5,67 +5,28 @@ import (
 	"strings"
 	"time"
 
-	"github.com/pocketbase/dbx"
 	"github.com/pocketbase/pocketbase/core"
 )
 
-type alertTask struct {
-	action      string // "schedule" or "cancel"
-	systemName  string
-	alertRecord *core.Record
-	delay       time.Duration
-}
-
 type alertInfo struct {
-	systemName  string
-	alertRecord *core.Record
-	expireTime  time.Time
+	systemName string
+	alertData  CachedAlertData
+	expireTime time.Time
+	timer      *time.Timer
 }
 
-// startWorker is a long-running goroutine that processes alert tasks
-// every x seconds. It must be running to process status alerts.
-func (am *AlertManager) startWorker() {
-	processPendingAlerts := time.Tick(15 * time.Second)
-
-	// check for status alerts that are not resolved when system comes up
-	// (can be removed if we figure out core bug in #1052)
-	checkStatusAlerts := time.Tick(561 * time.Second)
-
-	for {
-		select {
-		case <-am.stopChan:
-			return
-		case task := <-am.alertQueue:
-			switch task.action {
-			case "schedule":
-				am.pendingAlerts.Store(task.alertRecord.Id, &alertInfo{
-					systemName:  task.systemName,
-					alertRecord: task.alertRecord,
-					expireTime:  time.Now().Add(task.delay),
-				})
-			case "cancel":
-				am.pendingAlerts.Delete(task.alertRecord.Id)
+// Stop cancels all pending status alert timers.
+func (am *AlertManager) Stop() {
+	am.stopOnce.Do(func() {
+		am.pendingAlerts.Range(func(key, value any) bool {
+			info := value.(*alertInfo)
+			if info.timer != nil {
+				info.timer.Stop()
 			}
-		case <-checkStatusAlerts:
-			resolveStatusAlerts(am.hub)
-		case <-processPendingAlerts:
-			// Check for expired alerts every tick
-			now := time.Now()
-			for key, value := range am.pendingAlerts.Range {
-				info := value.(*alertInfo)
-				if now.After(info.expireTime) {
-					// Downtime delay has passed, process alert
-					am.sendStatusAlert("down", info.systemName, info.alertRecord)
-					am.pendingAlerts.Delete(key)
-				}
-			}
-		}
-	}
-}
-
-// StopWorker shuts down the AlertManager.worker goroutine
-func (am *AlertManager) StopWorker() {
-	close(am.stopChan)
+			am.pendingAlerts.Delete(key)
+			return true
+		})
+	})
 }
 
 // HandleStatusAlerts manages the logic when system status changes.
@@ -74,82 +35,116 @@ func (am *AlertManager) HandleStatusAlerts(newStatus string, systemRecord *core.
 		return nil
 	}
 
-	alertRecords, err := am.getSystemStatusAlerts(systemRecord.Id)
-	if err != nil {
-		return err
-	}
-	if len(alertRecords) == 0 {
+	alerts := am.alertsCache.GetAlertsByName(systemRecord.Id, "Status")
+	if len(alerts) == 0 {
 		return nil
 	}
 
 	systemName := systemRecord.GetString("name")
 	if newStatus == "down" {
-		am.handleSystemDown(systemName, alertRecords)
+		am.handleSystemDown(systemName, alerts)
 	} else {
-		am.handleSystemUp(systemName, alertRecords)
+		am.handleSystemUp(systemName, alerts)
 	}
 	return nil
 }
 
-// getSystemStatusAlerts retrieves all "Status" alert records for a given system ID.
-func (am *AlertManager) getSystemStatusAlerts(systemID string) ([]*core.Record, error) {
-	alertRecords, err := am.hub.FindAllRecords("alerts", dbx.HashExp{
-		"system": systemID,
-		"name":   "Status",
-	})
-	if err != nil {
-		return nil, err
+// handleSystemDown manages the logic when a system status changes to "down". It schedules pending alerts for each alert record.
+func (am *AlertManager) handleSystemDown(systemName string, alerts []CachedAlertData) {
+	for _, alertData := range alerts {
+		min := max(1, int(alertData.Min))
+		am.schedulePendingStatusAlert(systemName, alertData, time.Duration(min)*time.Minute)
 	}
-	return alertRecords, nil
 }
 
-// Schedules delayed "down" alerts for each alert record.
-func (am *AlertManager) handleSystemDown(systemName string, alertRecords []*core.Record) {
-	for _, alertRecord := range alertRecords {
-		// Continue if alert is already scheduled
-		if _, exists := am.pendingAlerts.Load(alertRecord.Id); exists {
-			continue
-		}
-		// Schedule by adding to queue
-		min := max(1, alertRecord.GetInt("min"))
-		am.alertQueue <- alertTask{
-			action:      "schedule",
-			systemName:  systemName,
-			alertRecord: alertRecord,
-			delay:       time.Duration(min) * time.Minute,
-		}
+// schedulePendingStatusAlert sets up a timer to send a "down" alert after the specified delay if the system is still down.
+// It returns true if the alert was scheduled, or false if an alert was already pending for the given alert record.
+func (am *AlertManager) schedulePendingStatusAlert(systemName string, alertData CachedAlertData, delay time.Duration) bool {
+	alert := &alertInfo{
+		systemName: systemName,
+		alertData:  alertData,
+		expireTime: time.Now().Add(delay),
 	}
+
+	storedAlert, loaded := am.pendingAlerts.LoadOrStore(alertData.Id, alert)
+	if loaded {
+		return false
+	}
+
+	stored := storedAlert.(*alertInfo)
+	stored.timer = time.AfterFunc(time.Until(stored.expireTime), func() {
+		am.processPendingAlert(alertData.Id)
+	})
+	return true
 }
 
 // handleSystemUp manages the logic when a system status changes to "up".
 // It cancels any pending alerts and sends "up" alerts.
-func (am *AlertManager) handleSystemUp(systemName string, alertRecords []*core.Record) {
-	for _, alertRecord := range alertRecords {
-		alertRecordID := alertRecord.Id
+func (am *AlertManager) handleSystemUp(systemName string, alerts []CachedAlertData) {
+	for _, alertData := range alerts {
 		// If alert exists for record, delete and continue (down alert not sent)
-		if _, exists := am.pendingAlerts.Load(alertRecordID); exists {
-			am.alertQueue <- alertTask{
-				action:      "cancel",
-				alertRecord: alertRecord,
-			}
+		if am.cancelPendingAlert(alertData.Id) {
 			continue
 		}
-		// No alert scheduled for this record, send "up" alert
-		if err := am.sendStatusAlert("up", systemName, alertRecord); err != nil {
+		if !alertData.Triggered {
+			continue
+		}
+		if err := am.sendStatusAlert("up", systemName, alertData); err != nil {
 			am.hub.Logger().Error("Failed to send alert", "err", err)
 		}
 	}
 }
 
-// sendStatusAlert sends a status alert ("up" or "down") to the users associated with the alert records.
-func (am *AlertManager) sendStatusAlert(alertStatus string, systemName string, alertRecord *core.Record) error {
-	switch alertStatus {
-	case "up":
-		alertRecord.Set("triggered", false)
-	case "down":
-		alertRecord.Set("triggered", true)
+// cancelPendingAlert stops the timer and removes the pending alert for the given alert ID. Returns true if a pending alert was found and cancelled.
+func (am *AlertManager) cancelPendingAlert(alertID string) bool {
+	value, loaded := am.pendingAlerts.LoadAndDelete(alertID)
+	if !loaded {
+		return false
+	}
+
+	info := value.(*alertInfo)
+	if info.timer != nil {
+		info.timer.Stop()
+	}
+	return true
+}
+
+// CancelPendingStatusAlerts cancels all pending status alert timers for a given system.
+// This is called when a system is paused to prevent delayed alerts from firing.
+func (am *AlertManager) CancelPendingStatusAlerts(systemID string) {
+	am.pendingAlerts.Range(func(key, value any) bool {
+		info := value.(*alertInfo)
+		if info.alertData.SystemID == systemID {
+			am.cancelPendingAlert(key.(string))
+		}
+		return true
+	})
+}
+
+// processPendingAlert sends a "down" alert if the pending alert has expired and the system is still down.
+func (am *AlertManager) processPendingAlert(alertID string) {
+	value, loaded := am.pendingAlerts.LoadAndDelete(alertID)
+	if !loaded {
+		return
+	}
+
+	info := value.(*alertInfo)
+	refreshedAlertData, ok := am.alertsCache.Refresh(info.alertData)
+	if !ok || refreshedAlertData.Triggered {
+		return
+	}
+	if err := am.sendStatusAlert("down", info.systemName, refreshedAlertData); err != nil {
+		am.hub.Logger().Error("Failed to send alert", "err", err)
+	}
+}
+
+// sendStatusAlert sends a status alert ("up" or "down") to the users associated with the alert records.
+func (am *AlertManager) sendStatusAlert(alertStatus string, systemName string, alertData CachedAlertData) error {
+	// Update trigger state for alert record before sending alert
+	triggered := alertStatus == "down"
+	if err := am.setAlertTriggered(alertData, triggered); err != nil {
+		return err
 	}
-	am.hub.Save(alertRecord)
 
 	var emoji string
 	if alertStatus == "up" {
@@ -162,10 +157,10 @@ func (am *AlertManager) sendStatusAlert(alertStatus string, systemName string, a
 	message := strings.TrimSuffix(title, emoji)
 
 	// Get system ID for the link
-	systemID := alertRecord.GetString("system")
+	systemID := alertData.SystemID
 
 	return am.SendAlert(AlertMessageData{
-		UserID:   alertRecord.GetString("user"),
+		UserID:   alertData.UserID,
 		SystemID: systemID,
 		Title:    title,
 		Message:  message,
@@ -174,8 +169,8 @@ func (am *AlertManager) sendStatusAlert(alertStatus string, systemName string, a
 	})
 }
 
-// resolveStatusAlerts resolves any status alerts that weren't resolved
-// when system came up (https://github.com/henrygd/beszel/issues/1052)
+// resolveStatusAlerts resolves any triggered status alerts that weren't resolved
+// when system came up (https://github.com/henrygd/beszel/issues/1052).
 func resolveStatusAlerts(app core.App) error {
 	db := app.DB()
 	// Find all active status alerts where the system is actually up
@@ -205,3 +200,40 @@ func resolveStatusAlerts(app core.App) error {
 	}
 	return nil
 }
+
+// restorePendingStatusAlerts re-queues untriggered status alerts for systems that
+// are still down after a hub restart. This rebuilds the lost in-memory timer state.
+func (am *AlertManager) restorePendingStatusAlerts() error {
+	type pendingStatusAlert struct {
+		AlertID    string `db:"alert_id"`
+		SystemID   string `db:"system_id"`
+		SystemName string `db:"system_name"`
+	}
+
+	var pending []pendingStatusAlert
+	err := am.hub.DB().NewQuery(`
+		SELECT a.id AS alert_id, a.system AS system_id, s.name AS system_name
+		FROM alerts a
+		JOIN systems s ON a.system = s.id
+		WHERE a.name = 'Status'
+		AND a.triggered = false
+		AND s.status = 'down'
+	`).All(&pending)
+	if err != nil {
+		return err
+	}
+
+	// Make sure cache is populated before trying to restore pending alerts
+	_ = am.alertsCache.PopulateFromDB(false)
+
+	for _, item := range pending {
+		alertData, ok := am.alertsCache.GetAlert(item.SystemID, item.AlertID)
+		if !ok {
+			continue
+		}
+		min := max(1, int(alertData.Min))
+		am.schedulePendingStatusAlert(item.SystemName, alertData, time.Duration(min)*time.Minute)
+	}
+
+	return nil
+}

internal/alerts/alerts_system.go

@@ -11,15 +11,11 @@ import (
 	"github.com/pocketbase/dbx"
 	"github.com/pocketbase/pocketbase/core"
 	"github.com/pocketbase/pocketbase/tools/types"
-	"github.com/spf13/cast"
 )
 
 func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *system.CombinedData) error {
-	alertRecords, err := am.hub.FindAllRecords("alerts",
-		dbx.NewExp("system={:system} AND name!='Status'", dbx.Params{"system": systemRecord.Id}),
-	)
-	if err != nil || len(alertRecords) == 0 {
-		// log.Println("no alerts found for system")
+	alerts := am.alertsCache.GetAlertsExcludingNames(systemRecord.Id, "Status")
+	if len(alerts) == 0 {
 		return nil
 	}
 
@@ -27,8 +23,8 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 	now := systemRecord.GetDateTime("updated").Time().UTC()
 	oldestTime := now
 
-	for _, alertRecord := range alertRecords {
-		name := alertRecord.GetString("name")
+	for _, alertData := range alerts {
+		name := alertData.Name
 		var val float64
 		unit := "%"
 
@@ -38,7 +34,7 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 		case "Memory":
 			val = data.Info.MemPct
 		case "Bandwidth":
-			val = data.Info.Bandwidth
+			val = float64(data.Info.BandwidthBytes) / (1024 * 1024)
 			unit = " MB/s"
 		case "Disk":
 			maxUsedPct := data.Info.DiskPct
@@ -73,8 +69,8 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 			val = float64(data.Stats.Battery[0])
 		}
 
-		triggered := alertRecord.GetBool("triggered")
-		threshold := alertRecord.GetFloat("value")
+		triggered := alertData.Triggered
+		threshold := alertData.Value
 
 		// Battery alert has inverted logic: trigger when value is BELOW threshold
 		lowAlert := isLowAlert(name)
@@ -92,11 +88,11 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 			}
 		}
 
-		min := max(1, cast.ToUint8(alertRecord.Get("min")))
+		min := max(1, alertData.Min)
 
 		alert := SystemAlertData{
 			systemRecord: systemRecord,
-			alertRecord:  alertRecord,
+			alertData:    alertData,
 			name:         name,
 			unit:         unit,
 			val:          val,
@@ -129,7 +125,7 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 		Created types.DateTime `db:"created"`
 	}{}
 
-	err = am.hub.DB().
+	err := am.hub.DB().
 		Select("stats", "created").
 		From("system_stats").
 		Where(dbx.NewExp(
@@ -192,22 +188,24 @@ func (am *AlertManager) HandleSystemAlerts(systemRecord *core.Record, data *syst
 			case "Memory":
 				alert.val += stats.Mem
 			case "Bandwidth":
-				alert.val += stats.NetSent + stats.NetRecv
+				alert.val += float64(stats.Bandwidth[0]+stats.Bandwidth[1]) / (1024 * 1024)
 			case "Disk":
 				if alert.mapSums == nil {
-					alert.mapSums = make(map[string]float32, len(data.Stats.ExtraFs)+1)
+					alert.mapSums = make(map[string]float32, len(stats.ExtraFs)+1)
 				}
 				// add root disk
 				if _, ok := alert.mapSums["root"]; !ok {
 					alert.mapSums["root"] = 0.0
 				}
 				alert.mapSums["root"] += float32(stats.Disk)
-				// add extra disks
-				for key, fs := range data.Stats.ExtraFs {
-					if _, ok := alert.mapSums[key]; !ok {
-						alert.mapSums[key] = 0.0
+				// add extra disks from historical record
+				for key, fs := range stats.ExtraFs {
+					if fs.DiskTotal > 0 {
+						if _, ok := alert.mapSums[key]; !ok {
+							alert.mapSums[key] = 0.0
+						}
+						alert.mapSums[key] += float32(fs.DiskUsed / fs.DiskTotal * 100)
 					}
-					alert.mapSums[key] += float32(fs.DiskUsed / fs.DiskTotal * 100)
 				}
 			case "Temperature":
 				if alert.mapSums == nil {
@@ -342,13 +340,12 @@ func (am *AlertManager) sendSystemAlert(alert SystemAlertData) {
 	}
 	body := fmt.Sprintf("%s averaged %.2f%s for the previous %v %s.", alert.descriptor, alert.val, alert.unit, alert.min, minutesLabel)
 
-	alert.alertRecord.Set("triggered", alert.triggered)
-	if err := am.hub.Save(alert.alertRecord); err != nil {
+	if err := am.setAlertTriggered(alert.alertData, alert.triggered); err != nil {
 		// app.Logger().Error("failed to save alert record", "err", err)
 		return
 	}
 	am.SendAlert(AlertMessageData{
-		UserID:   alert.alertRecord.GetString("user"),
+		UserID:   alert.alertData.UserID,
 		SystemID: alert.systemRecord.Id,
 		Title:    subject,
 		Message:  body,

internal/alerts/alerts_system_test.go

@@ -0,0 +1,218 @@
+//go:build testing
+
+package alerts_test
+
+import (
+	"testing"
+	"testing/synctest"
+	"time"
+
+	"github.com/henrygd/beszel/internal/entities/system"
+	beszelTests "github.com/henrygd/beszel/internal/tests"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+type systemAlertValueSetter[T any] func(info *system.Info, stats *system.Stats, value T)
+
+type systemAlertTestFixture struct {
+	hub     *beszelTests.TestHub
+	alertID string
+	submit  func(*system.CombinedData) error
+}
+
+func createCombinedData[T any](value T, setValue systemAlertValueSetter[T]) *system.CombinedData {
+	var data system.CombinedData
+	setValue(&data.Info, &data.Stats, value)
+	return &data
+}
+
+func newSystemAlertTestFixture(t *testing.T, alertName string, min int, threshold float64) *systemAlertTestFixture {
+	t.Helper()
+
+	hub, user := beszelTests.GetHubWithUser(t)
+
+	systems, err := beszelTests.CreateSystems(hub, 1, user.Id, "up")
+	require.NoError(t, err)
+	systemRecord := systems[0]
+
+	sysManagerSystem, err := hub.GetSystemManager().GetSystemFromStore(systemRecord.Id)
+	require.NoError(t, err)
+	require.NotNil(t, sysManagerSystem)
+	sysManagerSystem.StopUpdater()
+
+	userSettings, err := hub.FindFirstRecordByFilter("user_settings", "user={:user}", map[string]any{"user": user.Id})
+	require.NoError(t, err)
+	userSettings.Set("settings", `{"emails":["test@example.com"],"webhooks":[]}`)
+	require.NoError(t, hub.Save(userSettings))
+
+	alertRecord, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":   alertName,
+		"system": systemRecord.Id,
+		"user":   user.Id,
+		"min":    min,
+		"value":  threshold,
+	})
+	require.NoError(t, err)
+
+	assert.False(t, alertRecord.GetBool("triggered"), "Alert should not be triggered initially")
+
+	alertsCache := hub.GetAlertManager().GetSystemAlertsCache()
+	cachedAlerts := alertsCache.GetAlertsExcludingNames(systemRecord.Id, "Status")
+	assert.Len(t, cachedAlerts, 1, "Alert should be in cache")
+
+	return &systemAlertTestFixture{
+		hub:     hub,
+		alertID: alertRecord.Id,
+		submit: func(data *system.CombinedData) error {
+			_, err := sysManagerSystem.CreateRecords(data)
+			return err
+		},
+	}
+}
+
+func (fixture *systemAlertTestFixture) cleanup() {
+	fixture.hub.Cleanup()
+}
+
+func submitValue[T any](fixture *systemAlertTestFixture, t *testing.T, value T, setValue systemAlertValueSetter[T]) {
+	t.Helper()
+	require.NoError(t, fixture.submit(createCombinedData(value, setValue)))
+}
+
+func (fixture *systemAlertTestFixture) assertTriggered(t *testing.T, triggered bool, message string) {
+	t.Helper()
+
+	alertRecord, err := fixture.hub.FindRecordById("alerts", fixture.alertID)
+	require.NoError(t, err)
+	assert.Equal(t, triggered, alertRecord.GetBool("triggered"), message)
+}
+
+func waitForSystemAlert(d time.Duration) {
+	time.Sleep(d)
+	synctest.Wait()
+}
+
+func testOneMinuteSystemAlert[T any](t *testing.T, alertName string, threshold float64, setValue systemAlertValueSetter[T], triggerValue, resolveValue T) {
+	t.Helper()
+
+	synctest.Test(t, func(t *testing.T) {
+		fixture := newSystemAlertTestFixture(t, alertName, 1, threshold)
+		defer fixture.cleanup()
+
+		submitValue(fixture, t, triggerValue, setValue)
+		waitForSystemAlert(time.Second)
+
+		fixture.assertTriggered(t, true, "Alert should be triggered")
+		assert.Equal(t, 1, fixture.hub.TestMailer.TotalSend(), "An email should have been sent")
+
+		submitValue(fixture, t, resolveValue, setValue)
+		waitForSystemAlert(time.Second)
+
+		fixture.assertTriggered(t, false, "Alert should be untriggered")
+		assert.Equal(t, 2, fixture.hub.TestMailer.TotalSend(), "A second email should have been sent for untriggering the alert")
+
+		waitForSystemAlert(time.Minute)
+	})
+}
+
+func testMultiMinuteSystemAlert[T any](t *testing.T, alertName string, threshold float64, min int, setValue systemAlertValueSetter[T], baselineValue, triggerValue, resolveValue T) {
+	t.Helper()
+
+	synctest.Test(t, func(t *testing.T) {
+		fixture := newSystemAlertTestFixture(t, alertName, min, threshold)
+		defer fixture.cleanup()
+
+		submitValue(fixture, t, baselineValue, setValue)
+		waitForSystemAlert(time.Minute + time.Second)
+		fixture.assertTriggered(t, false, "Alert should not be triggered yet")
+
+		submitValue(fixture, t, triggerValue, setValue)
+		waitForSystemAlert(time.Minute)
+		fixture.assertTriggered(t, false, "Alert should not be triggered until the history window is full")
+
+		submitValue(fixture, t, triggerValue, setValue)
+		waitForSystemAlert(time.Second)
+		fixture.assertTriggered(t, true, "Alert should be triggered")
+		assert.Equal(t, 1, fixture.hub.TestMailer.TotalSend(), "An email should have been sent")
+
+		submitValue(fixture, t, resolveValue, setValue)
+		waitForSystemAlert(time.Second)
+		fixture.assertTriggered(t, false, "Alert should be untriggered")
+		assert.Equal(t, 2, fixture.hub.TestMailer.TotalSend(), "A second email should have been sent for untriggering the alert")
+	})
+}
+
+func setCPUAlertValue(info *system.Info, stats *system.Stats, value float64) {
+	info.Cpu = value
+	stats.Cpu = value
+}
+
+func setMemoryAlertValue(info *system.Info, stats *system.Stats, value float64) {
+	info.MemPct = value
+	stats.MemPct = value
+}
+
+func setDiskAlertValue(info *system.Info, stats *system.Stats, value float64) {
+	info.DiskPct = value
+	stats.DiskPct = value
+}
+
+func setBandwidthAlertValue(info *system.Info, stats *system.Stats, value [2]uint64) {
+	info.BandwidthBytes = value[0] + value[1]
+	stats.Bandwidth = value
+}
+
+func megabytesToBytes(mb uint64) uint64 {
+	return mb * 1024 * 1024
+}
+
+func setGPUAlertValue(info *system.Info, stats *system.Stats, value float64) {
+	info.GpuPct = value
+	stats.GPUData = map[string]system.GPUData{
+		"GPU0": {Usage: value},
+	}
+}
+
+func setTemperatureAlertValue(info *system.Info, stats *system.Stats, value float64) {
+	info.DashboardTemp = value
+	stats.Temperatures = map[string]float64{
+		"Temp0": value,
+	}
+}
+
+func setLoadAvgAlertValue(info *system.Info, stats *system.Stats, value [3]float64) {
+	info.LoadAvg = value
+	stats.LoadAvg = value
+}
+
+func setBatteryAlertValue(info *system.Info, stats *system.Stats, value [2]uint8) {
+	info.Battery = value
+	stats.Battery = value
+}
+
+func TestSystemAlertsOneMin(t *testing.T) {
+	testOneMinuteSystemAlert(t, "CPU", 50, setCPUAlertValue, 51, 49)
+	testOneMinuteSystemAlert(t, "Memory", 50, setMemoryAlertValue, 51, 49)
+	testOneMinuteSystemAlert(t, "Disk", 50, setDiskAlertValue, 51, 49)
+	testOneMinuteSystemAlert(t, "Bandwidth", 50, setBandwidthAlertValue, [2]uint64{megabytesToBytes(26), megabytesToBytes(25)}, [2]uint64{megabytesToBytes(25), megabytesToBytes(24)})
+	testOneMinuteSystemAlert(t, "GPU", 50, setGPUAlertValue, 51, 49)
+	testOneMinuteSystemAlert(t, "Temperature", 70, setTemperatureAlertValue, 71, 69)
+	testOneMinuteSystemAlert(t, "LoadAvg1", 4, setLoadAvgAlertValue, [3]float64{4.1, 0, 0}, [3]float64{3.9, 0, 0})
+	testOneMinuteSystemAlert(t, "LoadAvg5", 4, setLoadAvgAlertValue, [3]float64{0, 4.1, 0}, [3]float64{0, 3.9, 0})
+	testOneMinuteSystemAlert(t, "LoadAvg15", 4, setLoadAvgAlertValue, [3]float64{0, 0, 4.1}, [3]float64{0, 0, 3.9})
+	testOneMinuteSystemAlert(t, "Battery", 20, setBatteryAlertValue, [2]uint8{19, 0}, [2]uint8{21, 0})
+}
+
+func TestSystemAlertsTwoMin(t *testing.T) {
+	testMultiMinuteSystemAlert(t, "CPU", 50, 2, setCPUAlertValue, 10, 51, 48)
+	testMultiMinuteSystemAlert(t, "Memory", 50, 2, setMemoryAlertValue, 10, 51, 48)
+	testMultiMinuteSystemAlert(t, "Disk", 50, 2, setDiskAlertValue, 10, 51, 48)
+	testMultiMinuteSystemAlert(t, "Bandwidth", 50, 2, setBandwidthAlertValue, [2]uint64{megabytesToBytes(10), megabytesToBytes(10)}, [2]uint64{megabytesToBytes(26), megabytesToBytes(25)}, [2]uint64{megabytesToBytes(10), megabytesToBytes(10)})
+	testMultiMinuteSystemAlert(t, "GPU", 50, 2, setGPUAlertValue, 10, 51, 48)
+	testMultiMinuteSystemAlert(t, "Temperature", 70, 2, setTemperatureAlertValue, 10, 71, 67)
+	testMultiMinuteSystemAlert(t, "LoadAvg1", 4, 2, setLoadAvgAlertValue, [3]float64{0, 0, 0}, [3]float64{4.1, 0, 0}, [3]float64{3.5, 0, 0})
+	testMultiMinuteSystemAlert(t, "LoadAvg5", 4, 2, setLoadAvgAlertValue, [3]float64{0, 2, 0}, [3]float64{0, 4.1, 0}, [3]float64{0, 3.5, 0})
+	testMultiMinuteSystemAlert(t, "LoadAvg15", 4, 2, setLoadAvgAlertValue, [3]float64{0, 0, 2}, [3]float64{0, 0, 4.1}, [3]float64{0, 0, 3.5})
+	testMultiMinuteSystemAlert(t, "Battery", 20, 2, setBatteryAlertValue, [2]uint8{21, 0}, [2]uint8{19, 0}, [2]uint8{25, 1})
+}

internal/alerts/alerts_test.go

@@ -1,456 +1,19 @@
 //go:build testing
-// +build testing
 
 package alerts_test
 
 import (
-	"bytes"
-	"encoding/json"
-	"io"
-	"net/http"
-	"strings"
 	"testing"
 	"testing/synctest"
 	"time"
 
-	"github.com/henrygd/beszel/internal/alerts"
 	beszelTests "github.com/henrygd/beszel/internal/tests"
 
+	"github.com/henrygd/beszel/internal/alerts"
 	"github.com/pocketbase/dbx"
-	"github.com/pocketbase/pocketbase/core"
-	pbTests "github.com/pocketbase/pocketbase/tests"
 	"github.com/stretchr/testify/assert"
 )
 
-// marshal to json and return an io.Reader (for use in ApiScenario.Body)
-func jsonReader(v any) io.Reader {
-	data, err := json.Marshal(v)
-	if err != nil {
-		panic(err)
-	}
-	return bytes.NewReader(data)
-}
-
-func TestUserAlertsApi(t *testing.T) {
-	hub, _ := beszelTests.NewTestHub(t.TempDir())
-	defer hub.Cleanup()
-
-	hub.StartHub()
-
-	user1, _ := beszelTests.CreateUser(hub, "alertstest@example.com", "password")
-	user1Token, _ := user1.NewAuthToken()
-
-	user2, _ := beszelTests.CreateUser(hub, "alertstest2@example.com", "password")
-	user2Token, _ := user2.NewAuthToken()
-
-	system1, _ := beszelTests.CreateRecord(hub, "systems", map[string]any{
-		"name":  "system1",
-		"users": []string{user1.Id},
-		"host":  "127.0.0.1",
-	})
-
-	system2, _ := beszelTests.CreateRecord(hub, "systems", map[string]any{
-		"name":  "system2",
-		"users": []string{user1.Id, user2.Id},
-		"host":  "127.0.0.2",
-	})
-
-	userRecords, _ := hub.CountRecords("users")
-	assert.EqualValues(t, 2, userRecords, "all users should be created")
-
-	systemRecords, _ := hub.CountRecords("systems")
-	assert.EqualValues(t, 2, systemRecords, "all systems should be created")
-
-	testAppFactory := func(t testing.TB) *pbTests.TestApp {
-		return hub.TestApp
-	}
-
-	scenarios := []beszelTests.ApiScenario{
-		// {
-		// 	Name:            "GET not implemented - returns index",
-		// 	Method:          http.MethodGet,
-		// 	URL:             "/api/beszel/user-alerts",
-		// 	ExpectedStatus:  200,
-		// 	ExpectedContent: []string{"<html ", "globalThis.BESZEL"},
-		// 	TestAppFactory:  testAppFactory,
-		// },
-		{
-			Name:            "POST no auth",
-			Method:          http.MethodPost,
-			URL:             "/api/beszel/user-alerts",
-			ExpectedStatus:  401,
-			ExpectedContent: []string{"requires valid"},
-			TestAppFactory:  testAppFactory,
-		},
-		{
-			Name:   "POST no body",
-			Method: http.MethodPost,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  400,
-			ExpectedContent: []string{"Bad data"},
-			TestAppFactory:  testAppFactory,
-		},
-		{
-			Name:   "POST bad data",
-			Method: http.MethodPost,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  400,
-			ExpectedContent: []string{"Bad data"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"invalidField": "this should cause validation error",
-				"threshold":    "not a number",
-			}),
-		},
-		{
-			Name:   "POST malformed JSON",
-			Method: http.MethodPost,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  400,
-			ExpectedContent: []string{"Bad data"},
-			TestAppFactory:  testAppFactory,
-			Body:            strings.NewReader(`{"alertType": "cpu", "threshold": 80, "enabled": true,}`),
-		},
-		{
-			Name:   "POST valid alert data multiple systems",
-			Method: http.MethodPost,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  200,
-			ExpectedContent: []string{"\"success\":true"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":      "CPU",
-				"value":     69,
-				"min":       9,
-				"systems":   []string{system1.Id, system2.Id},
-				"overwrite": false,
-			}),
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				// check total alerts
-				alerts, _ := app.CountRecords("alerts")
-				assert.EqualValues(t, 2, alerts, "should have 2 alerts")
-				// check alert has correct values
-				matchingAlerts, _ := app.CountRecords("alerts", dbx.HashExp{"name": "CPU", "user": user1.Id, "system": system1.Id, "value": 69, "min": 9})
-				assert.EqualValues(t, 1, matchingAlerts, "should have 1 alert")
-			},
-		},
-		{
-			Name:   "POST valid alert data single system",
-			Method: http.MethodPost,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  200,
-			ExpectedContent: []string{"\"success\":true"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":    "Memory",
-				"systems": []string{system1.Id},
-				"value":   90,
-				"min":     10,
-			}),
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				user1Alerts, _ := app.CountRecords("alerts", dbx.HashExp{"user": user1.Id})
-				assert.EqualValues(t, 3, user1Alerts, "should have 3 alerts")
-			},
-		},
-		{
-			Name:   "Overwrite: false, should not overwrite existing alert",
-			Method: http.MethodPost,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  200,
-			ExpectedContent: []string{"\"success\":true"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":      "CPU",
-				"value":     45,
-				"min":       5,
-				"systems":   []string{system1.Id},
-				"overwrite": false,
-			}),
-			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
-				beszelTests.ClearCollection(t, app, "alerts")
-				beszelTests.CreateRecord(app, "alerts", map[string]any{
-					"name":   "CPU",
-					"system": system1.Id,
-					"user":   user1.Id,
-					"value":  80,
-					"min":    10,
-				})
-			},
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				alerts, _ := app.CountRecords("alerts")
-				assert.EqualValues(t, 1, alerts, "should have 1 alert")
-				alert, _ := app.FindFirstRecordByFilter("alerts", "name = 'CPU' && user = {:user}", dbx.Params{"user": user1.Id})
-				assert.EqualValues(t, 80, alert.Get("value"), "should have 80 as value")
-			},
-		},
-		{
-			Name:   "Overwrite: true, should overwrite existing alert",
-			Method: http.MethodPost,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user2Token,
-			},
-			ExpectedStatus:  200,
-			ExpectedContent: []string{"\"success\":true"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":      "CPU",
-				"value":     45,
-				"min":       5,
-				"systems":   []string{system2.Id},
-				"overwrite": true,
-			}),
-			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
-				beszelTests.ClearCollection(t, app, "alerts")
-				beszelTests.CreateRecord(app, "alerts", map[string]any{
-					"name":   "CPU",
-					"system": system2.Id,
-					"user":   user2.Id,
-					"value":  80,
-					"min":    10,
-				})
-			},
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				alerts, _ := app.CountRecords("alerts")
-				assert.EqualValues(t, 1, alerts, "should have 1 alert")
-				alert, _ := app.FindFirstRecordByFilter("alerts", "name = 'CPU' && user = {:user}", dbx.Params{"user": user2.Id})
-				assert.EqualValues(t, 45, alert.Get("value"), "should have 45 as value")
-			},
-		},
-		{
-			Name:            "DELETE no auth",
-			Method:          http.MethodDelete,
-			URL:             "/api/beszel/user-alerts",
-			ExpectedStatus:  401,
-			ExpectedContent: []string{"requires valid"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":    "CPU",
-				"systems": []string{system1.Id},
-			}),
-			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
-				beszelTests.ClearCollection(t, app, "alerts")
-				beszelTests.CreateRecord(app, "alerts", map[string]any{
-					"name":   "CPU",
-					"system": system1.Id,
-					"user":   user1.Id,
-					"value":  80,
-					"min":    10,
-				})
-			},
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				alerts, _ := app.CountRecords("alerts")
-				assert.EqualValues(t, 1, alerts, "should have 1 alert")
-			},
-		},
-		{
-			Name:   "DELETE alert",
-			Method: http.MethodDelete,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  200,
-			ExpectedContent: []string{"\"count\":1", "\"success\":true"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":    "CPU",
-				"systems": []string{system1.Id},
-			}),
-			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
-				beszelTests.ClearCollection(t, app, "alerts")
-				beszelTests.CreateRecord(app, "alerts", map[string]any{
-					"name":   "CPU",
-					"system": system1.Id,
-					"user":   user1.Id,
-					"value":  80,
-					"min":    10,
-				})
-			},
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				alerts, _ := app.CountRecords("alerts")
-				assert.Zero(t, alerts, "should have 0 alerts")
-			},
-		},
-		{
-			Name:   "DELETE alert multiple systems",
-			Method: http.MethodDelete,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user1Token,
-			},
-			ExpectedStatus:  200,
-			ExpectedContent: []string{"\"count\":2", "\"success\":true"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":    "Memory",
-				"systems": []string{system1.Id, system2.Id},
-			}),
-			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
-				beszelTests.ClearCollection(t, app, "alerts")
-				for _, systemId := range []string{system1.Id, system2.Id} {
-					_, err := beszelTests.CreateRecord(app, "alerts", map[string]any{
-						"name":   "Memory",
-						"system": systemId,
-						"user":   user1.Id,
-						"value":  90,
-						"min":    10,
-					})
-					assert.NoError(t, err, "should create alert")
-				}
-				alerts, _ := app.CountRecords("alerts")
-				assert.EqualValues(t, 2, alerts, "should have 2 alerts")
-			},
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				alerts, _ := app.CountRecords("alerts")
-				assert.Zero(t, alerts, "should have 0 alerts")
-			},
-		},
-		{
-			Name:   "User 2 should not be able to delete alert of user 1",
-			Method: http.MethodDelete,
-			URL:    "/api/beszel/user-alerts",
-			Headers: map[string]string{
-				"Authorization": user2Token,
-			},
-			ExpectedStatus:  200,
-			ExpectedContent: []string{"\"count\":1", "\"success\":true"},
-			TestAppFactory:  testAppFactory,
-			Body: jsonReader(map[string]any{
-				"name":    "CPU",
-				"systems": []string{system2.Id},
-			}),
-			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
-				beszelTests.ClearCollection(t, app, "alerts")
-				for _, user := range []string{user1.Id, user2.Id} {
-					beszelTests.CreateRecord(app, "alerts", map[string]any{
-						"name":   "CPU",
-						"system": system2.Id,
-						"user":   user,
-						"value":  80,
-						"min":    10,
-					})
-				}
-				alerts, _ := app.CountRecords("alerts")
-				assert.EqualValues(t, 2, alerts, "should have 2 alerts")
-				user1AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user1.Id})
-				assert.EqualValues(t, 1, user1AlertCount, "should have 1 alert")
-				user2AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user2.Id})
-				assert.EqualValues(t, 1, user2AlertCount, "should have 1 alert")
-			},
-			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
-				user1AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user1.Id})
-				assert.EqualValues(t, 1, user1AlertCount, "should have 1 alert")
-				user2AlertCount, _ := app.CountRecords("alerts", dbx.HashExp{"user": user2.Id})
-				assert.Zero(t, user2AlertCount, "should have 0 alerts")
-			},
-		},
-	}
-
-	for _, scenario := range scenarios {
-		scenario.Test(t)
-	}
-}
-
-func TestStatusAlerts(t *testing.T) {
-	synctest.Test(t, func(t *testing.T) {
-		hub, user := beszelTests.GetHubWithUser(t)
-		defer hub.Cleanup()
-
-		systems, err := beszelTests.CreateSystems(hub, 4, user.Id, "paused")
-		assert.NoError(t, err)
-
-		var alerts []*core.Record
-		for i, system := range systems {
-			alert, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
-				"name":   "Status",
-				"system": system.Id,
-				"user":   user.Id,
-				"min":    i + 1,
-			})
-			assert.NoError(t, err)
-			alerts = append(alerts, alert)
-		}
-
-		time.Sleep(10 * time.Millisecond)
-
-		for _, alert := range alerts {
-			assert.False(t, alert.GetBool("triggered"), "Alert should not be triggered immediately")
-		}
-		if hub.TestMailer.TotalSend() != 0 {
-			assert.Zero(t, hub.TestMailer.TotalSend(), "Expected 0 messages, got %d", hub.TestMailer.TotalSend())
-		}
-		for _, system := range systems {
-			assert.EqualValues(t, "paused", system.GetString("status"), "System should be paused")
-		}
-		for _, system := range systems {
-			system.Set("status", "up")
-			err = hub.SaveNoValidate(system)
-			assert.NoError(t, err)
-		}
-		time.Sleep(time.Second)
-		assert.EqualValues(t, 0, hub.GetPendingAlertsCount(), "should have 0 alerts in the pendingAlerts map")
-		for _, system := range systems {
-			system.Set("status", "down")
-			err = hub.SaveNoValidate(system)
-			assert.NoError(t, err)
-		}
-		// after 30 seconds, should have 4 alerts in the pendingAlerts map, no triggered alerts
-		time.Sleep(time.Second * 30)
-		assert.EqualValues(t, 4, hub.GetPendingAlertsCount(), "should have 4 alerts in the pendingAlerts map")
-		triggeredCount, err := hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
-		assert.NoError(t, err)
-		assert.EqualValues(t, 0, triggeredCount, "should have 0 alert triggered")
-		assert.EqualValues(t, 0, hub.TestMailer.TotalSend(), "should have 0 messages sent")
-		// after 1:30 seconds, should have 1 triggered alert and 3 pending alerts
-		time.Sleep(time.Second * 60)
-		assert.EqualValues(t, 3, hub.GetPendingAlertsCount(), "should have 3 alerts in the pendingAlerts map")
-		triggeredCount, err = hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
-		assert.NoError(t, err)
-		assert.EqualValues(t, 1, triggeredCount, "should have 1 alert triggered")
-		assert.EqualValues(t, 1, hub.TestMailer.TotalSend(), "should have 1 messages sent")
-		// after 2:30 seconds, should have 2 triggered alerts and 2 pending alerts
-		time.Sleep(time.Second * 60)
-		assert.EqualValues(t, 2, hub.GetPendingAlertsCount(), "should have 2 alerts in the pendingAlerts map")
-		triggeredCount, err = hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
-		assert.NoError(t, err)
-		assert.EqualValues(t, 2, triggeredCount, "should have 2 alert triggered")
-		assert.EqualValues(t, 2, hub.TestMailer.TotalSend(), "should have 2 messages sent")
-		// now we will bring the remaning systems back up
-		for _, system := range systems {
-			system.Set("status", "up")
-			err = hub.SaveNoValidate(system)
-			assert.NoError(t, err)
-		}
-		time.Sleep(time.Second)
-		// should have 0 alerts in the pendingAlerts map and 0 alerts triggered
-		assert.EqualValues(t, 0, hub.GetPendingAlertsCount(), "should have 0 alerts in the pendingAlerts map")
-		triggeredCount, err = hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
-		assert.NoError(t, err)
-		assert.Zero(t, triggeredCount, "should have 0 alert triggered")
-		// 4 messages sent, 2 down alerts and 2 up alerts for first 2 systems
-		assert.EqualValues(t, 4, hub.TestMailer.TotalSend(), "should have 4 messages sent")
-	})
-}
-
 func TestAlertsHistory(t *testing.T) {
 	synctest.Test(t, func(t *testing.T) {
 		hub, user := beszelTests.GetHubWithUser(t)
@@ -579,102 +142,46 @@ func TestAlertsHistory(t *testing.T) {
 		assert.EqualValues(t, 2, totalHistoryCount, "Should have 2 total alert history records")
 	})
 }
-func TestResolveStatusAlerts(t *testing.T) {
-	hub, user := beszelTests.GetHubWithUser(t)
+
+func TestSetAlertTriggered(t *testing.T) {
+	hub, _ := beszelTests.NewTestHub(t.TempDir())
 	defer hub.Cleanup()
 
-	// Create a systemUp
-	systemUp, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
-		"name":   "test-system",
-		"users":  []string{user.Id},
-		"host":   "127.0.0.1",
-		"status": "up",
+	hub.StartHub()
+
+	user, _ := beszelTests.CreateUser(hub, "test@example.com", "password")
+	system, _ := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "test-system",
+		"users": []string{user.Id},
+		"host":  "127.0.0.1",
 	})
-	assert.NoError(t, err)
 
-	systemDown, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
-		"name":   "test-system-2",
-		"users":  []string{user.Id},
-		"host":   "127.0.0.2",
-		"status": "up",
+	alertRecord, _ := beszelTests.CreateRecord(hub, "alerts", map[string]any{
+		"name":      "CPU",
+		"system":    system.Id,
+		"user":      user.Id,
+		"value":     80,
+		"triggered": false,
 	})
+
+	am := alerts.NewAlertManager(hub)
+
+	var alert alerts.CachedAlertData
+	alert.PopulateFromRecord(alertRecord)
+
+	// Test triggering the alert
+	err := am.SetAlertTriggered(alert, true)
 	assert.NoError(t, err)
 
-	// Create a status alertUp for the system
-	alertUp, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
-		"name":   "Status",
-		"system": systemUp.Id,
-		"user":   user.Id,
-		"min":    1,
-	})
+	updatedRecord, err := hub.FindRecordById("alerts", alert.Id)
+	assert.NoError(t, err)
+	assert.True(t, updatedRecord.GetBool("triggered"))
+
+	// Test un-triggering the alert
+	err = am.SetAlertTriggered(alert, false)
 	assert.NoError(t, err)
 
-	alertDown, err := beszelTests.CreateRecord(hub, "alerts", map[string]any{
-		"name":   "Status",
-		"system": systemDown.Id,
-		"user":   user.Id,
-		"min":    1,
-	})
+	updatedRecord, err = hub.FindRecordById("alerts", alert.Id)
 	assert.NoError(t, err)
-
-	// Verify alert is not triggered initially
-	assert.False(t, alertUp.GetBool("triggered"), "Alert should not be triggered initially")
-
-	// Set the system to 'up' (this should not trigger the alert)
-	systemUp.Set("status", "up")
-	err = hub.SaveNoValidate(systemUp)
-	assert.NoError(t, err)
-
-	systemDown.Set("status", "down")
-	err = hub.SaveNoValidate(systemDown)
-	assert.NoError(t, err)
-
-	// Wait a moment for any processing
-	time.Sleep(10 * time.Millisecond)
-
-	// Verify alertUp is still not triggered after setting system to up
-	alertUp, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertUp.Id})
-	assert.NoError(t, err)
-	assert.False(t, alertUp.GetBool("triggered"), "Alert should not be triggered when system is up")
-
-	// Manually set both alerts triggered to true
-	alertUp.Set("triggered", true)
-	err = hub.SaveNoValidate(alertUp)
-	assert.NoError(t, err)
-	alertDown.Set("triggered", true)
-	err = hub.SaveNoValidate(alertDown)
-	assert.NoError(t, err)
-
-	// Verify we have exactly one alert with triggered true
-	triggeredCount, err := hub.CountRecords("alerts", dbx.HashExp{"triggered": true})
-	assert.NoError(t, err)
-	assert.EqualValues(t, 2, triggeredCount, "Should have exactly two alerts with triggered true")
-
-	// Verify the specific alertUp is triggered
-	alertUp, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertUp.Id})
-	assert.NoError(t, err)
-	assert.True(t, alertUp.GetBool("triggered"), "Alert should be triggered")
-
-	// Verify we have two unresolved alert history records
-	alertHistoryCount, err := hub.CountRecords("alerts_history", dbx.HashExp{"resolved": ""})
-	assert.NoError(t, err)
-	assert.EqualValues(t, 2, alertHistoryCount, "Should have exactly two unresolved alert history records")
-
-	err = alerts.ResolveStatusAlerts(hub)
-	assert.NoError(t, err)
-
-	// Verify alertUp is not triggered after resolving
-	alertUp, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertUp.Id})
-	assert.NoError(t, err)
-	assert.False(t, alertUp.GetBool("triggered"), "Alert should not be triggered after resolving")
-	// Verify alertDown is still triggered
-	alertDown, err = hub.FindFirstRecordByFilter("alerts", "id={:id}", dbx.Params{"id": alertDown.Id})
-	assert.NoError(t, err)
-	assert.True(t, alertDown.GetBool("triggered"), "Alert should still be triggered after resolving")
-
-	// Verify we have one unresolved alert history record
-	alertHistoryCount, err = hub.CountRecords("alerts_history", dbx.HashExp{"resolved": ""})
-	assert.NoError(t, err)
-	assert.EqualValues(t, 1, alertHistoryCount, "Should have exactly one unresolved alert history record")
-
+	assert.False(t, updatedRecord.GetBool("triggered"))
 }

internal/alerts/alerts_test_helpers.go

@@ -1,5 +1,4 @@
 //go:build testing
-// +build testing
 
 package alerts
 
@@ -10,6 +9,18 @@ import (
 	"github.com/pocketbase/pocketbase/core"
 )
 
+func NewTestAlertManagerWithoutWorker(app hubLike) *AlertManager {
+	return &AlertManager{
+		hub:         app,
+		alertsCache: NewAlertsCache(app),
+	}
+}
+
+// GetSystemAlertsCache returns the internal system alerts cache.
+func (am *AlertManager) GetSystemAlertsCache() *AlertsCache {
+	return am.alertsCache
+}
+
 func (am *AlertManager) GetAlertManager() *AlertManager {
 	return am
 }
@@ -28,19 +39,18 @@ func (am *AlertManager) GetPendingAlertsCount() int {
 }
 
 // ProcessPendingAlerts manually processes all expired alerts (for testing)
-func (am *AlertManager) ProcessPendingAlerts() ([]*core.Record, error) {
+func (am *AlertManager) ProcessPendingAlerts() ([]CachedAlertData, error) {
 	now := time.Now()
 	var lastErr error
-	var processedAlerts []*core.Record
+	var processedAlerts []CachedAlertData
 	am.pendingAlerts.Range(func(key, value any) bool {
 		info := value.(*alertInfo)
 		if now.After(info.expireTime) {
-			// Downtime delay has passed, process alert
-			if err := am.sendStatusAlert("down", info.systemName, info.alertRecord); err != nil {
-				lastErr = err
+			if info.timer != nil {
+				info.timer.Stop()
 			}
-			processedAlerts = append(processedAlerts, info.alertRecord)
-			am.pendingAlerts.Delete(key)
+			am.processPendingAlert(key.(string))
+			processedAlerts = append(processedAlerts, info.alertData)
 		}
 		return true
 	})
@@ -57,6 +67,35 @@ func (am *AlertManager) ForceExpirePendingAlerts() {
 	})
 }
 
+func (am *AlertManager) ResetPendingAlertTimer(alertID string, delay time.Duration) bool {
+	value, loaded := am.pendingAlerts.Load(alertID)
+	if !loaded {
+		return false
+	}
+
+	info := value.(*alertInfo)
+	if info.timer != nil {
+		info.timer.Stop()
+	}
+	info.expireTime = time.Now().Add(delay)
+	info.timer = time.AfterFunc(delay, func() {
+		am.processPendingAlert(alertID)
+	})
+	return true
+}
+
 func ResolveStatusAlerts(app core.App) error {
 	return resolveStatusAlerts(app)
 }
+
+func (am *AlertManager) RestorePendingStatusAlerts() error {
+	return am.restorePendingStatusAlerts()
+}
+
+func (am *AlertManager) SetAlertTriggered(alert CachedAlertData, triggered bool) error {
+	return am.setAlertTriggered(alert, triggered)
+}
+
+func IsInternalURL(rawURL string) (bool, error) {
+	return isInternalURL(rawURL)
+}

internal/cmd/agent/agent.go

@@ -9,6 +9,7 @@ import (
 	"github.com/henrygd/beszel"
 	"github.com/henrygd/beszel/agent"
 	"github.com/henrygd/beszel/agent/health"
+	"github.com/henrygd/beszel/agent/utils"
 	"github.com/spf13/pflag"
 	"golang.org/x/crypto/ssh"
 )
@@ -31,9 +32,6 @@ func (opts *cmdOptions) parse() bool {
 
 	// Subcommands that don't require any pflag parsing
 	switch subcommand {
-	case "-v", "version":
-		fmt.Println(beszel.AppName+"-agent", beszel.Version)
-		return true
 	case "health":
 		err := health.Check()
 		if err != nil {
@@ -41,6 +39,9 @@ func (opts *cmdOptions) parse() bool {
 		}
 		fmt.Print("ok")
 		return true
+	case "fingerprint":
+		handleFingerprint()
+		return true
 	}
 
 	// pflag.CommandLine.ParseErrorsWhitelist.UnknownFlags = true
@@ -49,6 +50,7 @@ func (opts *cmdOptions) parse() bool {
 	pflag.StringVarP(&opts.hubURL, "url", "u", "", "URL of the Beszel hub")
 	pflag.StringVarP(&opts.token, "token", "t", "", "Token to use for authentication")
 	chinaMirrors := pflag.BoolP("china-mirrors", "c", false, "Use mirror for update (gh.beszel.dev) instead of GitHub")
+	version := pflag.BoolP("version", "v", false, "Show version information")
 	help := pflag.BoolP("help", "h", false, "Show this help message")
 
 	// Convert old single-dash long flags to double-dash for backward compatibility
@@ -73,9 +75,9 @@ func (opts *cmdOptions) parse() bool {
 		builder.WriteString(os.Args[0])
 		builder.WriteString(" [command] [flags]\n")
 		builder.WriteString("\nCommands:\n")
-		builder.WriteString("  health    Check if the agent is running\n")
-		// builder.WriteString("  help      Display this help message\n")
-		builder.WriteString("  update    Update to the latest version\n")
+		builder.WriteString("  fingerprint  View or reset the agent fingerprint\n")
+		builder.WriteString("  health       Check if the agent is running\n")
+		builder.WriteString("  update       Update to the latest version\n")
 		builder.WriteString("\nFlags:\n")
 		fmt.Print(builder.String())
 		pflag.PrintDefaults()
@@ -86,6 +88,9 @@ func (opts *cmdOptions) parse() bool {
 
 	// Must run after pflag.Parse()
 	switch {
+	case *version:
+		fmt.Println(beszel.AppName+"-agent", beszel.Version)
+		return true
 	case *help || subcommand == "help":
 		pflag.Usage()
 		return true
@@ -112,12 +117,12 @@ func (opts *cmdOptions) loadPublicKeys() ([]ssh.PublicKey, error) {
 	}
 
 	// Try environment variable
-	if key, ok := agent.GetEnv("KEY"); ok && key != "" {
+	if key, ok := utils.GetEnv("KEY"); ok && key != "" {
 		return agent.ParseKeys(key)
 	}
 
 	// Try key file
-	keyFile, ok := agent.GetEnv("KEY_FILE")
+	keyFile, ok := utils.GetEnv("KEY_FILE")
 	if !ok {
 		return nil, fmt.Errorf("no key provided: must set -key flag, KEY env var, or KEY_FILE env var. Use 'beszel-agent help' for usage")
 	}
@@ -133,6 +138,38 @@ func (opts *cmdOptions) getAddress() string {
 	return agent.GetAddress(opts.listen)
 }
 
+// handleFingerprint handles the "fingerprint" command with subcommands "view" and "reset".
+func handleFingerprint() {
+	subCmd := ""
+	if len(os.Args) > 2 {
+		subCmd = os.Args[2]
+	}
+
+	switch subCmd {
+	case "", "view":
+		dataDir, _ := agent.GetDataDir()
+		fp := agent.GetFingerprint(dataDir, "", "")
+		fmt.Println(fp)
+	case "help", "-h", "--help":
+		fmt.Print(fingerprintUsage())
+	case "reset":
+		dataDir, err := agent.GetDataDir()
+		if err != nil {
+			log.Fatal(err)
+		}
+		if err := agent.DeleteFingerprint(dataDir); err != nil {
+			log.Fatal(err)
+		}
+		fmt.Println("Fingerprint reset. A new one will be generated on next start.")
+	default:
+		log.Fatalf("Unknown command: %q\n\n%s", subCmd, fingerprintUsage())
+	}
+}
+
+func fingerprintUsage() string {
+	return fmt.Sprintf("Usage: %s fingerprint [view|reset]\n\nCommands:\n  view   Print fingerprint (default)\n  reset  Reset saved fingerprint\n", os.Args[0])
+}
+
 func main() {
 	var opts cmdOptions
 	subcommandHandled := opts.parse()
@@ -158,6 +195,6 @@ func main() {
 	}
 
 	if err := a.Start(serverConfig); err != nil {
-		log.Fatal("Failed to start server: ", err)
+		log.Fatal("Failed to start: ", err)
 	}
 }

internal/cmd/hub/hub.go

@@ -28,8 +28,8 @@ func main() {
 	}
 
 	baseApp := getBaseApp()
-	h := hub.NewHub(baseApp)
-	if err := h.StartHub(); err != nil {
+	hub := hub.NewHub(baseApp)
+	if err := hub.StartHub(); err != nil {
 		log.Fatal(err)
 	}
 }

internal/dockerfile_agent

@@ -23,6 +23,9 @@ COPY --from=builder /agent /agent
 # this is so we don't need to create the /tmp directory in the scratch container
 COPY --from=builder /tmp /tmp
 
+# AMD GPU name lookup (used by agent on Linux when /usr/share/libdrm/amdgpu.ids is read)
+COPY --from=builder /app/agent/test-data/amdgpu.ids /usr/share/libdrm/amdgpu.ids
+
 # Ensure data persistence across container recreations
 VOLUME ["/var/lib/beszel-agent"]
 

internal/dockerfile_agent_alpine

@@ -20,6 +20,9 @@ RUN rm -rf /tmp/*
 FROM alpine:3.23
 COPY --from=builder /agent /agent
 
+# AMD GPU name lookup (used by agent on Linux when /usr/share/libdrm/amdgpu.ids is read)
+COPY --from=builder /app/agent/test-data/amdgpu.ids /usr/share/libdrm/amdgpu.ids
+
 RUN apk add --no-cache smartmontools
 
 # Ensure data persistence across container recreations

internal/dockerfile_agent_nvidia

@@ -37,6 +37,9 @@ RUN apt-get update && apt-get install -y \
 FROM nvidia/cuda:12.2.2-base-ubuntu22.04
 COPY --from=builder /agent /agent
 
+# AMD GPU name lookup (used by agent on hybrid laptops when /usr/share/libdrm/amdgpu.ids is read)
+COPY --from=builder /app/agent/test-data/amdgpu.ids /usr/share/libdrm/amdgpu.ids
+
 # Copy smartmontools binaries and config files
 COPY --from=smartmontools-builder /usr/sbin/smartctl /usr/sbin/smartctl
 

internal/entities/container/container.go

@@ -10,10 +10,19 @@ type ApiInfo struct {
 	Status  string
 	State   string
 	Image   string
+	Health  struct {
+		Status string
+		// FailingStreak int
+	}
+	Ports []struct {
+		// PrivatePort uint16
+		PublicPort uint16
+		IP         string
+		// Type        string
+	}
 	// ImageID string
 	// Command string
 	// Created int64
-	// Ports      []Port
 	// SizeRw     int64 `json:",omitempty"`
 	// SizeRootFs int64 `json:",omitempty"`
 	// Labels     map[string]string
@@ -140,6 +149,7 @@ type Stats struct {
 	Status string       `json:"-" cbor:"6,keyasint"`
 	Id     string       `json:"-" cbor:"7,keyasint"`
 	Image  string       `json:"-" cbor:"8,keyasint"`
+	Ports  string       `json:"-" cbor:"10,keyasint"`
 	// PrevCpu     [2]uint64    `json:"-"`
 	CpuSystem    uint64       `json:"-"`
 	CpuContainer uint64       `json:"-"`

internal/entities/smart/smart.go

@@ -143,8 +143,8 @@ type AtaDeviceStatisticsPage struct {
 }
 
 type AtaDeviceStatisticsEntry struct {
-	Name  string  `json:"name"`
-	Value *uint64 `json:"value,omitempty"`
+	Name  string `json:"name"`
+	Value *int64 `json:"value,omitempty"`
 }
 
 type AtaSmartAttribute struct {
@@ -356,8 +356,8 @@ type SmartInfoForSata struct {
 	SmartStatus SmartStatusInfo `json:"smart_status"`
 	// AtaSmartData                 AtaSmartData                 `json:"ata_smart_data"`
 	// AtaSctCapabilities           AtaSctCapabilities           `json:"ata_sct_capabilities"`
-	AtaSmartAttributes  AtaSmartAttributes  `json:"ata_smart_attributes"`
-	AtaDeviceStatistics AtaDeviceStatistics `json:"ata_device_statistics"`
+	AtaSmartAttributes  AtaSmartAttributes `json:"ata_smart_attributes"`
+	AtaDeviceStatistics json.RawMessage    `json:"ata_device_statistics"`
 	// PowerOnTime                  PowerOnTimeInfo              `json:"power_on_time"`
 	// PowerCycleCount              uint16                       `json:"power_cycle_count"`
 	Temperature TemperatureInfo `json:"temperature"`
@@ -494,7 +494,7 @@ type SmartInfoForNvme struct {
 	FirmwareVersion string           `json:"firmware_version"`
 	// NVMePCIVendor                 NVMePCIVendor                 `json:"nvme_pci_vendor"`
 	// NVMeIEEEOUIIdentifier         uint32                        `json:"nvme_ieee_oui_identifier"`
-	// NVMeTotalCapacity             uint64                        `json:"nvme_total_capacity"`
+	NVMeTotalCapacity             uint64                        `json:"nvme_total_capacity"`
 	// NVMeUnallocatedCapacity       uint64                        `json:"nvme_unallocated_capacity"`
 	// NVMeControllerID              uint16                        `json:"nvme_controller_id"`
 	// NVMeVersion                   VersionStringInfo             `json:"nvme_version"`

internal/entities/system/system.go

@@ -12,8 +12,9 @@ import (
 
 type Stats struct {
 	Cpu            float64             `json:"cpu" cbor:"0,keyasint"`
-	MaxCpu         float64             `json:"cpum,omitempty" cbor:"1,keyasint,omitempty"`
+	MaxCpu         float64             `json:"cpum,omitempty" cbor:"-"`
 	Mem            float64             `json:"m" cbor:"2,keyasint"`
+	MaxMem         float64             `json:"mm,omitempty" cbor:"-"`
 	MemUsed        float64             `json:"mu" cbor:"3,keyasint"`
 	MemPct         float64             `json:"mp" cbor:"4,keyasint"`
 	MemBuffCache   float64             `json:"mb" cbor:"5,keyasint"`
@@ -23,31 +24,32 @@ type Stats struct {
 	DiskTotal      float64             `json:"d" cbor:"9,keyasint"`
 	DiskUsed       float64             `json:"du" cbor:"10,keyasint"`
 	DiskPct        float64             `json:"dp" cbor:"11,keyasint"`
-	DiskReadPs     float64             `json:"dr" cbor:"12,keyasint"`
-	DiskWritePs    float64             `json:"dw" cbor:"13,keyasint"`
-	MaxDiskReadPs  float64             `json:"drm,omitempty" cbor:"14,keyasint,omitempty"`
-	MaxDiskWritePs float64             `json:"dwm,omitempty" cbor:"15,keyasint,omitempty"`
+	DiskReadPs     float64             `json:"dr,omitzero" cbor:"12,keyasint,omitzero"`
+	DiskWritePs    float64             `json:"dw,omitzero" cbor:"13,keyasint,omitzero"`
+	MaxDiskReadPs  float64             `json:"drm,omitempty" cbor:"-"`
+	MaxDiskWritePs float64             `json:"dwm,omitempty" cbor:"-"`
 	NetworkSent    float64             `json:"ns,omitzero" cbor:"16,keyasint,omitzero"`
 	NetworkRecv    float64             `json:"nr,omitzero" cbor:"17,keyasint,omitzero"`
-	MaxNetworkSent float64             `json:"nsm,omitempty" cbor:"18,keyasint,omitempty"`
-	MaxNetworkRecv float64             `json:"nrm,omitempty" cbor:"19,keyasint,omitempty"`
+	MaxNetworkSent float64             `json:"nsm,omitempty" cbor:"-"`
+	MaxNetworkRecv float64             `json:"nrm,omitempty" cbor:"-"`
 	Temperatures   map[string]float64  `json:"t,omitempty" cbor:"20,keyasint,omitempty"`
 	ExtraFs        map[string]*FsStats `json:"efs,omitempty" cbor:"21,keyasint,omitempty"`
 	GPUData        map[string]GPUData  `json:"g,omitempty" cbor:"22,keyasint,omitempty"`
-	LoadAvg1       float64             `json:"l1,omitempty" cbor:"23,keyasint,omitempty"`
-	LoadAvg5       float64             `json:"l5,omitempty" cbor:"24,keyasint,omitempty"`
-	LoadAvg15      float64             `json:"l15,omitempty" cbor:"25,keyasint,omitempty"`
-	Bandwidth      [2]uint64           `json:"b,omitzero" cbor:"26,keyasint,omitzero"`  // [sent bytes, recv bytes]
-	MaxBandwidth   [2]uint64           `json:"bm,omitzero" cbor:"27,keyasint,omitzero"` // [sent bytes, recv bytes]
+	// LoadAvg1       float64             `json:"l1,omitempty" cbor:"23,keyasint,omitempty"`
+	// LoadAvg5       float64             `json:"l5,omitempty" cbor:"24,keyasint,omitempty"`
+	// LoadAvg15      float64             `json:"l15,omitempty" cbor:"25,keyasint,omitempty"`
+	Bandwidth    [2]uint64 `json:"b,omitzero" cbor:"26,keyasint,omitzero"` // [sent bytes, recv bytes]
+	MaxBandwidth [2]uint64 `json:"bm,omitzero" cbor:"-"`                   // [sent bytes, recv bytes]
 	// TODO: remove other load fields in future release in favor of load avg array
 	LoadAvg           [3]float64           `json:"la,omitempty" cbor:"28,keyasint"`
-	Battery           [2]uint8             `json:"bat,omitzero" cbor:"29,keyasint,omitzero"` // [percent, charge state, current]
-	MaxMem            float64              `json:"mm,omitempty" cbor:"30,keyasint,omitempty"`
+	Battery           [2]uint8             `json:"bat,omitzero" cbor:"29,keyasint,omitzero"`    // [percent, charge state, current]
 	NetworkInterfaces map[string][4]uint64 `json:"ni,omitempty" cbor:"31,keyasint,omitempty"`   // [upload bytes, download bytes, total upload, total download]
 	DiskIO            [2]uint64            `json:"dio,omitzero" cbor:"32,keyasint,omitzero"`    // [read bytes, write bytes]
 	MaxDiskIO         [2]uint64            `json:"diom,omitzero" cbor:"-"`                      // [max read bytes, max write bytes]
 	CpuBreakdown      []float64            `json:"cpub,omitempty" cbor:"33,keyasint,omitempty"` // [user, system, iowait, steal, idle]
 	CpuCoresUsage     Uint8Slice           `json:"cpus,omitempty" cbor:"34,keyasint,omitempty"` // per-core busy usage [CPU0..]
+	DiskIoStats       [6]float64           `json:"dios,omitzero" cbor:"35,keyasint,omitzero"`   // [read time %, write time %, io utilization %, r_await ms, w_await ms, weighted io %]
+	MaxDiskIoStats    [6]float64           `json:"diosm,omitzero" cbor:"-"`                     // max values for DiskIoStats
 }
 
 // Uint8Slice wraps []uint8 to customize JSON encoding while keeping CBOR efficient.
@@ -90,13 +92,15 @@ type FsStats struct {
 	TotalWrite     uint64    `json:"-"`
 	DiskReadPs     float64   `json:"r" cbor:"2,keyasint"`
 	DiskWritePs    float64   `json:"w" cbor:"3,keyasint"`
-	MaxDiskReadPS  float64   `json:"rm,omitempty" cbor:"4,keyasint,omitempty"`
-	MaxDiskWritePS float64   `json:"wm,omitempty" cbor:"5,keyasint,omitempty"`
+	MaxDiskReadPS  float64   `json:"rm,omitempty" cbor:"-"`
+	MaxDiskWritePS float64   `json:"wm,omitempty" cbor:"-"`
 	// TODO: remove DiskReadPs and DiskWritePs in future release in favor of DiskReadBytes and DiskWriteBytes
-	DiskReadBytes     uint64 `json:"rb" cbor:"6,keyasint,omitempty"`
-	DiskWriteBytes    uint64 `json:"wb" cbor:"7,keyasint,omitempty"`
-	MaxDiskReadBytes  uint64 `json:"rbm,omitempty" cbor:"-"`
-	MaxDiskWriteBytes uint64 `json:"wbm,omitempty" cbor:"-"`
+	DiskReadBytes     uint64     `json:"rb" cbor:"6,keyasint,omitempty"`
+	DiskWriteBytes    uint64     `json:"wb" cbor:"7,keyasint,omitempty"`
+	MaxDiskReadBytes  uint64     `json:"rbm,omitempty" cbor:"-"`
+	MaxDiskWriteBytes uint64     `json:"wbm,omitempty" cbor:"-"`
+	DiskIoStats       [6]float64 `json:"dios,omitzero" cbor:"8,keyasint,omitzero"` // [read time %, write time %, io utilization %, r_await ms, w_await ms, weighted io %]
+	MaxDiskIoStats    [6]float64 `json:"diosm,omitzero" cbor:"-"`                  // max values for DiskIoStats
 }
 
 type NetIoStats struct {
@@ -129,23 +133,23 @@ type Info struct {
 	KernelVersion string `json:"k,omitempty" cbor:"1,keyasint,omitempty"` // deprecated - moved to Details struct
 	Cores         int    `json:"c,omitzero" cbor:"2,keyasint,omitzero"`   // deprecated - moved to Details struct
 	// Threads is needed in Info struct to calculate load average thresholds
-	Threads        int     `json:"t,omitempty" cbor:"3,keyasint,omitempty"`
-	CpuModel       string  `json:"m,omitempty" cbor:"4,keyasint,omitempty"` // deprecated - moved to Details struct
-	Uptime         uint64  `json:"u" cbor:"5,keyasint"`
-	Cpu            float64 `json:"cpu" cbor:"6,keyasint"`
-	MemPct         float64 `json:"mp" cbor:"7,keyasint"`
-	DiskPct        float64 `json:"dp" cbor:"8,keyasint"`
-	Bandwidth      float64 `json:"b" cbor:"9,keyasint"`
-	AgentVersion   string  `json:"v" cbor:"10,keyasint"`
-	Podman         bool    `json:"p,omitempty" cbor:"11,keyasint,omitempty"` // deprecated - moved to Details struct
-	GpuPct         float64 `json:"g,omitempty" cbor:"12,keyasint,omitempty"`
-	DashboardTemp  float64 `json:"dt,omitempty" cbor:"13,keyasint,omitempty"`
-	Os             Os      `json:"os,omitempty" cbor:"14,keyasint,omitempty"`  // deprecated - moved to Details struct
-	LoadAvg1       float64 `json:"l1,omitempty" cbor:"15,keyasint,omitempty"`  // deprecated - use `la` array instead
-	LoadAvg5       float64 `json:"l5,omitempty" cbor:"16,keyasint,omitempty"`  // deprecated - use `la` array instead
-	LoadAvg15      float64 `json:"l15,omitempty" cbor:"17,keyasint,omitempty"` // deprecated - use `la` array instead
-	BandwidthBytes uint64  `json:"bb" cbor:"18,keyasint"`
+	Threads       int     `json:"t,omitempty" cbor:"3,keyasint,omitempty"`
+	CpuModel      string  `json:"m,omitempty" cbor:"4,keyasint,omitempty"` // deprecated - moved to Details struct
+	Uptime        uint64  `json:"u" cbor:"5,keyasint"`
+	Cpu           float64 `json:"cpu" cbor:"6,keyasint"`
+	MemPct        float64 `json:"mp" cbor:"7,keyasint"`
+	DiskPct       float64 `json:"dp" cbor:"8,keyasint"`
+	Bandwidth     float64 `json:"b,omitzero" cbor:"9,keyasint"` // deprecated in favor of BandwidthBytes
+	AgentVersion  string  `json:"v" cbor:"10,keyasint"`
+	Podman        bool    `json:"p,omitempty" cbor:"11,keyasint,omitempty"` // deprecated - moved to Details struct
+	GpuPct        float64 `json:"g,omitempty" cbor:"12,keyasint,omitempty"`
+	DashboardTemp float64 `json:"dt,omitempty" cbor:"13,keyasint,omitempty"`
+	Os            Os      `json:"os,omitempty" cbor:"14,keyasint,omitempty"` // deprecated - moved to Details struct
+	// LoadAvg1       float64 `json:"l1,omitempty" cbor:"15,keyasint,omitempty"`  // deprecated - use `la` array instead
+	// LoadAvg5       float64 `json:"l5,omitempty" cbor:"16,keyasint,omitempty"`  // deprecated - use `la` array instead
+	// LoadAvg15      float64 `json:"l15,omitempty" cbor:"17,keyasint,omitempty"` // deprecated - use `la` array instead
 
+	BandwidthBytes uint64             `json:"bb" cbor:"18,keyasint"`
 	LoadAvg        [3]float64         `json:"la,omitempty" cbor:"19,keyasint"`
 	ConnectionType ConnectionType     `json:"ct,omitempty" cbor:"20,keyasint,omitempty,omitzero"`
 	ExtraFsPct     map[string]float64 `json:"efs,omitempty" cbor:"21,keyasint,omitempty"`

internal/ghupdate/ghupdate.go

@@ -34,7 +34,7 @@ func ColorPrint(color, text string) {
 	fmt.Println(color + text + colorReset)
 }
 
-func ColorPrintf(color, format string, args ...interface{}) {
+func ColorPrintf(color, format string, args ...any) {
 	fmt.Printf(color+format+colorReset+"\n", args...)
 }
 
@@ -110,21 +110,13 @@ func (p *updater) update() (updated bool, err error) {
 	}
 
 	var latest *release
-	var useMirror bool
 
-	// Determine the API endpoint based on UseMirror flag
-	apiURL := fmt.Sprintf("https://api.github.com/repos/%s/%s/releases/latest", p.config.Owner, p.config.Repo)
+	apiURL := getApiURL(p.config.UseMirror, p.config.Owner, p.config.Repo)
 	if p.config.UseMirror {
-		useMirror = true
-		apiURL = fmt.Sprintf("https://gh.beszel.dev/repos/%s/%s/releases/latest?api=true", p.config.Owner, p.config.Repo)
 		ColorPrint(ColorYellow, "Using mirror for update.")
 	}
 
-	latest, err = fetchLatestRelease(
-		p.config.Context,
-		p.config.HttpClient,
-		apiURL,
-	)
+	latest, err = FetchLatestRelease(p.config.Context, p.config.HttpClient, apiURL)
 	if err != nil {
 		return false, err
 	}
@@ -150,7 +142,7 @@ func (p *updater) update() (updated bool, err error) {
 
 	// download the release asset
 	assetPath := filepath.Join(releaseDir, asset.Name)
-	if err := downloadFile(p.config.Context, p.config.HttpClient, asset.DownloadUrl, assetPath, useMirror); err != nil {
+	if err := downloadFile(p.config.Context, p.config.HttpClient, asset.DownloadUrl, assetPath, p.config.UseMirror); err != nil {
 		return false, err
 	}
 
@@ -226,11 +218,11 @@ func (p *updater) update() (updated bool, err error) {
 	return true, nil
 }
 
-func fetchLatestRelease(
-	ctx context.Context,
-	client HttpClient,
-	url string,
-) (*release, error) {
+func FetchLatestRelease(ctx context.Context, client HttpClient, url string) (*release, error) {
+	if url == "" {
+		url = getApiURL(false, "henrygd", "beszel")
+	}
+
 	req, err := http.NewRequestWithContext(ctx, "GET", url, nil)
 	if err != nil {
 		return nil, err
@@ -375,3 +367,10 @@ func isGlibc() bool {
 	}
 	return false
 }
+
+func getApiURL(useMirror bool, owner, repo string) string {
+	if useMirror {
+		return fmt.Sprintf("https://gh.beszel.dev/repos/%s/%s/releases/latest?api=true", owner, repo)
+	}
+	return fmt.Sprintf("https://api.github.com/repos/%s/%s/releases/latest", owner, repo)
+}

internal/hub/agent_connect_test.go

@@ -1,5 +1,4 @@
 //go:build testing
-// +build testing
 
 package hub
 
@@ -10,6 +9,7 @@ import (
 	"net/http/httptest"
 	"os"
 	"path/filepath"
+	"runtime"
 	"strings"
 	"testing"
 	"time"
@@ -32,7 +32,27 @@ func createTestHub(t testing.TB) (*Hub, *pbtests.TestApp, error) {
 	if err != nil {
 		return nil, nil, err
 	}
-	return NewHub(testApp), testApp, nil
+	return NewHub(testApp), testApp, err
+}
+
+// cleanupTestHub stops background system goroutines before tearing down the app.
+func cleanupTestHub(hub *Hub, testApp *pbtests.TestApp) {
+	if hub != nil {
+		sm := hub.GetSystemManager()
+		sm.RemoveAllSystems()
+		// Give updater goroutines a brief window to observe cancellation before DB teardown.
+		for range 20 {
+			if sm.GetSystemCount() == 0 {
+				break
+			}
+			runtime.Gosched()
+			time.Sleep(5 * time.Millisecond)
+		}
+		time.Sleep(20 * time.Millisecond)
+	}
+	if testApp != nil {
+		testApp.Cleanup()
+	}
 }
 
 // Helper function to create a test record
@@ -64,7 +84,7 @@ func TestValidateAgentHeaders(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	defer testApp.Cleanup()
+	defer cleanupTestHub(hub, testApp)
 
 	testCases := []struct {
 		name          string
@@ -145,7 +165,7 @@ func TestGetAllFingerprintRecordsByToken(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	defer testApp.Cleanup()
+	defer cleanupTestHub(hub, testApp)
 
 	// create test user
 	userRecord, err := createTestUser(testApp)
@@ -235,7 +255,7 @@ func TestSetFingerprint(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	defer testApp.Cleanup()
+	defer cleanupTestHub(hub, testApp)
 
 	// Create test user
 	userRecord, err := createTestUser(testApp)
@@ -315,7 +335,7 @@ func TestCreateSystemFromAgentData(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	defer testApp.Cleanup()
+	defer cleanupTestHub(hub, testApp)
 
 	// Create test user
 	userRecord, err := createTestUser(testApp)
@@ -425,7 +445,7 @@ func TestUniversalTokenFlow(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	defer testApp.Cleanup()
+	defer cleanupTestHub(nil, testApp)
 
 	// Create test user
 	userRecord, err := createTestUser(testApp)
@@ -493,7 +513,7 @@ func TestAgentConnect(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	defer testApp.Cleanup()
+	defer cleanupTestHub(hub, testApp)
 
 	// Create test user
 	userRecord, err := createTestUser(testApp)
@@ -652,7 +672,7 @@ func TestHandleAgentConnect(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	defer testApp.Cleanup()
+	defer cleanupTestHub(hub, testApp)
 
 	// Create test user
 	userRecord, err := createTestUser(testApp)
@@ -737,7 +757,7 @@ func TestAgentWebSocketIntegration(t *testing.T) {
 	// Create hub and test app
 	hub, testApp, err := createTestHub(t)
 	require.NoError(t, err)
-	defer testApp.Cleanup()
+	defer cleanupTestHub(hub, testApp)
 
 	// Get the hub's SSH key
 	hubSigner, err := hub.GetSSHKey("")
@@ -877,12 +897,8 @@ func TestAgentWebSocketIntegration(t *testing.T) {
 			require.NoError(t, err)
 
 			// Set up environment variables for the agent
-			os.Setenv("BESZEL_AGENT_HUB_URL", ts.URL)
-			os.Setenv("BESZEL_AGENT_TOKEN", tc.agentToken)
-			defer func() {
-				os.Unsetenv("BESZEL_AGENT_HUB_URL")
-				os.Unsetenv("BESZEL_AGENT_TOKEN")
-			}()
+			t.Setenv("BESZEL_AGENT_HUB_URL", ts.URL)
+			t.Setenv("BESZEL_AGENT_TOKEN", tc.agentToken)
 
 			// Start agent in background
 			done := make(chan error, 1)
@@ -897,7 +913,7 @@ func TestAgentWebSocketIntegration(t *testing.T) {
 
 			// Wait for connection result
 			maxWait := 2 * time.Second
-			time.Sleep(20 * time.Millisecond)
+			time.Sleep(40 * time.Millisecond)
 			checkInterval := 20 * time.Millisecond
 			timeout := time.After(maxWait)
 			ticker := time.Tick(checkInterval)
@@ -942,6 +958,8 @@ func TestAgentWebSocketIntegration(t *testing.T) {
 				}
 			}
 
+			time.Sleep(20 * time.Millisecond)
+
 			// Verify fingerprint state by re-reading the specific record
 			updatedFingerprintRecord, err := testApp.FindRecordById("fingerprints", fingerprintRecord.Id)
 			require.NoError(t, err)
@@ -976,7 +994,7 @@ func TestMultipleSystemsWithSameUniversalToken(t *testing.T) {
 	// Create hub and test app
 	hub, testApp, err := createTestHub(t)
 	require.NoError(t, err)
-	defer testApp.Cleanup()
+	defer cleanupTestHub(hub, testApp)
 
 	// Get the hub's SSH key
 	hubSigner, err := hub.GetSSHKey("")
@@ -1058,12 +1076,8 @@ func TestMultipleSystemsWithSameUniversalToken(t *testing.T) {
 			require.NoError(t, err)
 
 			// Set up environment variables for the agent
-			os.Setenv("BESZEL_AGENT_HUB_URL", ts.URL)
-			os.Setenv("BESZEL_AGENT_TOKEN", universalToken)
-			defer func() {
-				os.Unsetenv("BESZEL_AGENT_HUB_URL")
-				os.Unsetenv("BESZEL_AGENT_TOKEN")
-			}()
+			t.Setenv("BESZEL_AGENT_HUB_URL", ts.URL)
+			t.Setenv("BESZEL_AGENT_TOKEN", universalToken)
 
 			// Count systems before connection
 			systemsBefore, err := testApp.FindRecordsByFilter("systems", "users ~ {:userId}", "", -1, 0, map[string]any{"userId": userRecord.Id})
@@ -1144,6 +1158,8 @@ func TestMultipleSystemsWithSameUniversalToken(t *testing.T) {
 					assert.Equal(t, systemCount, systemsAfterCount, "Total system count should remain the same")
 				}
 
+				time.Sleep(20 * time.Millisecond)
+
 				// Verify that a fingerprint record exists for this fingerprint
 				fingerprints, err := testApp.FindRecordsByFilter("fingerprints", "token = {:token} && fingerprint = {:fingerprint}", "", -1, 0, map[string]any{
 					"token":       universalToken,
@@ -1176,7 +1192,7 @@ func TestPermanentUniversalTokenFromDB(t *testing.T) {
 	// Create hub and test app
 	hub, testApp, err := createTestHub(t)
 	require.NoError(t, err)
-	defer testApp.Cleanup()
+	defer cleanupTestHub(hub, testApp)
 
 	// Get the hub's SSH key
 	hubSigner, err := hub.GetSSHKey("")
@@ -1219,12 +1235,8 @@ func TestPermanentUniversalTokenFromDB(t *testing.T) {
 	require.NoError(t, err)
 
 	// Set up environment variables for the agent
-	os.Setenv("BESZEL_AGENT_HUB_URL", ts.URL)
-	os.Setenv("BESZEL_AGENT_TOKEN", universalToken)
-	defer func() {
-		os.Unsetenv("BESZEL_AGENT_HUB_URL")
-		os.Unsetenv("BESZEL_AGENT_TOKEN")
-	}()
+	t.Setenv("BESZEL_AGENT_HUB_URL", ts.URL)
+	t.Setenv("BESZEL_AGENT_TOKEN", universalToken)
 
 	// Start agent in background
 	done := make(chan error, 1)
@@ -1273,7 +1285,7 @@ verify:
 func TestFindOrCreateSystemForToken(t *testing.T) {
 	hub, testApp, err := createTestHub(t)
 	require.NoError(t, err)
-	defer testApp.Cleanup()
+	defer cleanupTestHub(hub, testApp)
 
 	// Create test user
 	userRecord, err := createTestUser(testApp)

internal/hub/api.go

@@ -0,0 +1,391 @@
+package hub
+
+import (
+	"context"
+	"net/http"
+	"regexp"
+	"strings"
+	"time"
+
+	"github.com/blang/semver"
+	"github.com/google/uuid"
+	"github.com/henrygd/beszel"
+	"github.com/henrygd/beszel/internal/alerts"
+	"github.com/henrygd/beszel/internal/ghupdate"
+	"github.com/henrygd/beszel/internal/hub/config"
+	"github.com/henrygd/beszel/internal/hub/systems"
+	"github.com/henrygd/beszel/internal/hub/utils"
+	"github.com/pocketbase/dbx"
+	"github.com/pocketbase/pocketbase/apis"
+	"github.com/pocketbase/pocketbase/core"
+)
+
+// UpdateInfo holds information about the latest update check
+type UpdateInfo struct {
+	lastCheck time.Time
+	Version   string `json:"v"`
+	Url       string `json:"url"`
+}
+
+var containerIDPattern = regexp.MustCompile(`^[a-fA-F0-9]{12,64}$`)
+
+// Middleware to allow only admin role users
+var requireAdminRole = customAuthMiddleware(func(e *core.RequestEvent) bool {
+	return e.Auth.GetString("role") == "admin"
+})
+
+// Middleware to exclude readonly users
+var excludeReadOnlyRole = customAuthMiddleware(func(e *core.RequestEvent) bool {
+	return e.Auth.GetString("role") != "readonly"
+})
+
+// customAuthMiddleware handles boilerplate for custom authentication middlewares. fn should
+// return true if the request is allowed, false otherwise. e.Auth is guaranteed to be non-nil.
+func customAuthMiddleware(fn func(*core.RequestEvent) bool) func(*core.RequestEvent) error {
+	return func(e *core.RequestEvent) error {
+		if e.Auth == nil {
+			return e.UnauthorizedError("The request requires valid record authorization token.", nil)
+		}
+		if !fn(e) {
+			return e.ForbiddenError("The authorized record is not allowed to perform this action.", nil)
+		}
+		return e.Next()
+	}
+}
+
+// registerMiddlewares registers custom middlewares
+func (h *Hub) registerMiddlewares(se *core.ServeEvent) {
+	// authorizes request with user matching the provided email
+	authorizeRequestWithEmail := func(e *core.RequestEvent, email string) (err error) {
+		if e.Auth != nil || email == "" {
+			return e.Next()
+		}
+		isAuthRefresh := e.Request.URL.Path == "/api/collections/users/auth-refresh" && e.Request.Method == http.MethodPost
+		e.Auth, err = e.App.FindAuthRecordByEmail("users", email)
+		if err != nil || !isAuthRefresh {
+			return e.Next()
+		}
+		// auth refresh endpoint, make sure token is set in header
+		token, _ := e.Auth.NewAuthToken()
+		e.Request.Header.Set("Authorization", token)
+		return e.Next()
+	}
+	// authenticate with trusted header
+	if autoLogin, _ := utils.GetEnv("AUTO_LOGIN"); autoLogin != "" {
+		se.Router.BindFunc(func(e *core.RequestEvent) error {
+			return authorizeRequestWithEmail(e, autoLogin)
+		})
+	}
+	// authenticate with trusted header
+	if trustedHeader, _ := utils.GetEnv("TRUSTED_AUTH_HEADER"); trustedHeader != "" {
+		se.Router.BindFunc(func(e *core.RequestEvent) error {
+			return authorizeRequestWithEmail(e, e.Request.Header.Get(trustedHeader))
+		})
+	}
+}
+
+// registerApiRoutes registers custom API routes
+func (h *Hub) registerApiRoutes(se *core.ServeEvent) error {
+	// auth protected routes
+	apiAuth := se.Router.Group("/api/beszel")
+	apiAuth.Bind(apis.RequireAuth())
+	// auth optional routes
+	apiNoAuth := se.Router.Group("/api/beszel")
+
+	// create first user endpoint only needed if no users exist
+	if totalUsers, _ := se.App.CountRecords("users"); totalUsers == 0 {
+		apiNoAuth.POST("/create-user", h.um.CreateFirstUser)
+	}
+	// check if first time setup on login page
+	apiNoAuth.GET("/first-run", func(e *core.RequestEvent) error {
+		total, err := e.App.CountRecords("users")
+		return e.JSON(http.StatusOK, map[string]bool{"firstRun": err == nil && total == 0})
+	})
+	// get public key and version
+	apiAuth.GET("/info", h.getInfo)
+	apiAuth.GET("/getkey", h.getInfo) // deprecated - keep for compatibility w/ integrations
+	// check for updates
+	if optIn, _ := utils.GetEnv("CHECK_UPDATES"); optIn == "true" {
+		var updateInfo UpdateInfo
+		apiAuth.GET("/update", updateInfo.getUpdate)
+	}
+	// send test notification
+	apiAuth.POST("/test-notification", h.SendTestNotification)
+	// heartbeat status and test
+	apiAuth.GET("/heartbeat-status", h.getHeartbeatStatus).BindFunc(requireAdminRole)
+	apiAuth.POST("/test-heartbeat", h.testHeartbeat).BindFunc(requireAdminRole)
+	// get config.yml content
+	apiAuth.GET("/config-yaml", config.GetYamlConfig).BindFunc(requireAdminRole)
+	// handle agent websocket connection
+	apiNoAuth.GET("/agent-connect", h.handleAgentConnect)
+	// get or create universal tokens
+	apiAuth.GET("/universal-token", h.getUniversalToken).BindFunc(excludeReadOnlyRole)
+	// update / delete user alerts
+	apiAuth.POST("/user-alerts", alerts.UpsertUserAlerts)
+	apiAuth.DELETE("/user-alerts", alerts.DeleteUserAlerts)
+	// refresh SMART devices for a system
+	apiAuth.POST("/smart/refresh", h.refreshSmartData).BindFunc(excludeReadOnlyRole)
+	// get systemd service details
+	apiAuth.GET("/systemd/info", h.getSystemdInfo)
+	// /containers routes
+	if enabled, _ := utils.GetEnv("CONTAINER_DETAILS"); enabled != "false" {
+		// get container logs
+		apiAuth.GET("/containers/logs", h.getContainerLogs)
+		// get container info
+		apiAuth.GET("/containers/info", h.getContainerInfo)
+	}
+	return nil
+}
+
+// getInfo returns data needed by authenticated users, such as the public key and current version
+func (h *Hub) getInfo(e *core.RequestEvent) error {
+	type infoResponse struct {
+		Key         string `json:"key"`
+		Version     string `json:"v"`
+		CheckUpdate bool   `json:"cu"`
+	}
+	info := infoResponse{
+		Key:     h.pubKey,
+		Version: beszel.Version,
+	}
+	if optIn, _ := utils.GetEnv("CHECK_UPDATES"); optIn == "true" {
+		info.CheckUpdate = true
+	}
+	return e.JSON(http.StatusOK, info)
+}
+
+// getUpdate checks for the latest release on GitHub and returns update info if a newer version is available
+func (info *UpdateInfo) getUpdate(e *core.RequestEvent) error {
+	if time.Since(info.lastCheck) < 6*time.Hour {
+		return e.JSON(http.StatusOK, info)
+	}
+	info.lastCheck = time.Now()
+	latestRelease, err := ghupdate.FetchLatestRelease(context.Background(), http.DefaultClient, "")
+	if err != nil {
+		return err
+	}
+	currentVersion, err := semver.Parse(strings.TrimPrefix(beszel.Version, "v"))
+	if err != nil {
+		return err
+	}
+	latestVersion, err := semver.Parse(strings.TrimPrefix(latestRelease.Tag, "v"))
+	if err != nil {
+		return err
+	}
+	if latestVersion.GT(currentVersion) {
+		info.Version = strings.TrimPrefix(latestRelease.Tag, "v")
+		info.Url = latestRelease.Url
+	}
+	return e.JSON(http.StatusOK, info)
+}
+
+// GetUniversalToken handles the universal token API endpoint (create, read, delete)
+func (h *Hub) getUniversalToken(e *core.RequestEvent) error {
+	if e.Auth.IsSuperuser() {
+		return e.ForbiddenError("Superusers cannot use universal tokens", nil)
+	}
+
+	tokenMap := universalTokenMap.GetMap()
+	userID := e.Auth.Id
+	query := e.Request.URL.Query()
+	token := query.Get("token")
+	enable := query.Get("enable")
+	permanent := query.Get("permanent")
+
+	// helper for deleting any existing permanent token record for this user
+	deletePermanent := func() error {
+		rec, err := h.FindFirstRecordByFilter("universal_tokens", "user = {:user}", dbx.Params{"user": userID})
+		if err != nil {
+			return nil // no record
+		}
+		return h.Delete(rec)
+	}
+
+	// helper for upserting a permanent token record for this user
+	upsertPermanent := func(token string) error {
+		rec, err := h.FindFirstRecordByFilter("universal_tokens", "user = {:user}", dbx.Params{"user": userID})
+		if err == nil {
+			rec.Set("token", token)
+			return h.Save(rec)
+		}
+
+		col, err := h.FindCachedCollectionByNameOrId("universal_tokens")
+		if err != nil {
+			return err
+		}
+		newRec := core.NewRecord(col)
+		newRec.Set("user", userID)
+		newRec.Set("token", token)
+		return h.Save(newRec)
+	}
+
+	// Disable universal tokens (both ephemeral and permanent)
+	if enable == "0" {
+		tokenMap.RemovebyValue(userID)
+		_ = deletePermanent()
+		return e.JSON(http.StatusOK, map[string]any{"token": token, "active": false, "permanent": false})
+	}
+
+	// Enable universal token (ephemeral or permanent)
+	if enable == "1" {
+		if token == "" {
+			token = uuid.New().String()
+		}
+
+		if permanent == "1" {
+			// make token permanent (persist across restarts)
+			tokenMap.RemovebyValue(userID)
+			if err := upsertPermanent(token); err != nil {
+				return err
+			}
+			return e.JSON(http.StatusOK, map[string]any{"token": token, "active": true, "permanent": true})
+		}
+
+		// default: ephemeral mode (1 hour)
+		_ = deletePermanent()
+		tokenMap.Set(token, userID, time.Hour)
+		return e.JSON(http.StatusOK, map[string]any{"token": token, "active": true, "permanent": false})
+	}
+
+	// Read current state
+	// Prefer permanent token if it exists.
+	if rec, err := h.FindFirstRecordByFilter("universal_tokens", "user = {:user}", dbx.Params{"user": userID}); err == nil {
+		dbToken := rec.GetString("token")
+		// If no token was provided, or the caller is asking about their permanent token, return it.
+		if token == "" || token == dbToken {
+			return e.JSON(http.StatusOK, map[string]any{"token": dbToken, "active": true, "permanent": true})
+		}
+		// Token doesn't match their permanent token (avoid leaking other info)
+		return e.JSON(http.StatusOK, map[string]any{"token": token, "active": false, "permanent": false})
+	}
+
+	// No permanent token; fall back to ephemeral token map.
+	if token == "" {
+		// return existing token if it exists
+		if token, _, ok := tokenMap.GetByValue(userID); ok {
+			return e.JSON(http.StatusOK, map[string]any{"token": token, "active": true, "permanent": false})
+		}
+		// if no token is provided, generate a new one
+		token = uuid.New().String()
+	}
+
+	// Token is considered active only if it belongs to the current user.
+	activeUser, ok := tokenMap.GetOk(token)
+	active := ok && activeUser == userID
+	response := map[string]any{"token": token, "active": active, "permanent": false}
+	return e.JSON(http.StatusOK, response)
+}
+
+// getHeartbeatStatus returns current heartbeat configuration and whether it's enabled
+func (h *Hub) getHeartbeatStatus(e *core.RequestEvent) error {
+	if h.hb == nil {
+		return e.JSON(http.StatusOK, map[string]any{
+			"enabled": false,
+			"msg":     "Set HEARTBEAT_URL to enable outbound heartbeat monitoring",
+		})
+	}
+	cfg := h.hb.GetConfig()
+	return e.JSON(http.StatusOK, map[string]any{
+		"enabled":  true,
+		"url":      cfg.URL,
+		"interval": cfg.Interval,
+		"method":   cfg.Method,
+	})
+}
+
+// testHeartbeat triggers a single heartbeat ping and returns the result
+func (h *Hub) testHeartbeat(e *core.RequestEvent) error {
+	if h.hb == nil {
+		return e.JSON(http.StatusOK, map[string]any{
+			"err": "Heartbeat not configured. Set HEARTBEAT_URL environment variable.",
+		})
+	}
+	if err := h.hb.Send(); err != nil {
+		return e.JSON(http.StatusOK, map[string]any{"err": err.Error()})
+	}
+	return e.JSON(http.StatusOK, map[string]any{"err": false})
+}
+
+// containerRequestHandler handles both container logs and info requests
+func (h *Hub) containerRequestHandler(e *core.RequestEvent, fetchFunc func(*systems.System, string) (string, error), responseKey string) error {
+	systemID := e.Request.URL.Query().Get("system")
+	containerID := e.Request.URL.Query().Get("container")
+
+	if systemID == "" || containerID == "" || !containerIDPattern.MatchString(containerID) {
+		return e.BadRequestError("Invalid system or container parameter", nil)
+	}
+
+	system, err := h.sm.GetSystem(systemID)
+	if err != nil || !system.HasUser(e.App, e.Auth) {
+		return e.NotFoundError("", nil)
+	}
+
+	data, err := fetchFunc(system, containerID)
+	if err != nil {
+		return e.InternalServerError("", err)
+	}
+
+	return e.JSON(http.StatusOK, map[string]string{responseKey: data})
+}
+
+// getContainerLogs handles GET /api/beszel/containers/logs requests
+func (h *Hub) getContainerLogs(e *core.RequestEvent) error {
+	return h.containerRequestHandler(e, func(system *systems.System, containerID string) (string, error) {
+		return system.FetchContainerLogsFromAgent(containerID)
+	}, "logs")
+}
+
+func (h *Hub) getContainerInfo(e *core.RequestEvent) error {
+	return h.containerRequestHandler(e, func(system *systems.System, containerID string) (string, error) {
+		return system.FetchContainerInfoFromAgent(containerID)
+	}, "info")
+}
+
+// getSystemdInfo handles GET /api/beszel/systemd/info requests
+func (h *Hub) getSystemdInfo(e *core.RequestEvent) error {
+	query := e.Request.URL.Query()
+	systemID := query.Get("system")
+	serviceName := query.Get("service")
+
+	if systemID == "" || serviceName == "" {
+		return e.BadRequestError("Invalid system or service parameter", nil)
+	}
+	system, err := h.sm.GetSystem(systemID)
+	if err != nil || !system.HasUser(e.App, e.Auth) {
+		return e.NotFoundError("", nil)
+	}
+	// verify service exists before fetching details
+	_, err = e.App.FindFirstRecordByFilter("systemd_services", "system = {:system} && name = {:name}", dbx.Params{
+		"system": systemID,
+		"name":   serviceName,
+	})
+	if err != nil {
+		return e.NotFoundError("", err)
+	}
+	details, err := system.FetchSystemdInfoFromAgent(serviceName)
+	if err != nil {
+		return e.InternalServerError("", err)
+	}
+	e.Response.Header().Set("Cache-Control", "public, max-age=60")
+	return e.JSON(http.StatusOK, map[string]any{"details": details})
+}
+
+// refreshSmartData handles POST /api/beszel/smart/refresh requests
+// Fetches fresh SMART data from the agent and updates the collection
+func (h *Hub) refreshSmartData(e *core.RequestEvent) error {
+	systemID := e.Request.URL.Query().Get("system")
+	if systemID == "" {
+		return e.BadRequestError("Invalid system parameter", nil)
+	}
+
+	system, err := h.sm.GetSystem(systemID)
+	if err != nil || !system.HasUser(e.App, e.Auth) {
+		return e.NotFoundError("", nil)
+	}
+
+	if err := system.FetchAndSaveSmartDevices(); err != nil {
+		return e.InternalServerError("", err)
+	}
+
+	return e.JSON(http.StatusOK, map[string]string{"status": "ok"})
+}

internal/hub/api_test.go

@@ -0,0 +1,972 @@
+package hub_test
+
+import (
+	"bytes"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"testing"
+
+	beszelTests "github.com/henrygd/beszel/internal/tests"
+
+	"github.com/henrygd/beszel/internal/migrations"
+	"github.com/pocketbase/pocketbase/core"
+	pbTests "github.com/pocketbase/pocketbase/tests"
+	"github.com/stretchr/testify/require"
+)
+
+// marshal to json and return an io.Reader (for use in ApiScenario.Body)
+func jsonReader(v any) io.Reader {
+	data, err := json.Marshal(v)
+	if err != nil {
+		panic(err)
+	}
+	return bytes.NewReader(data)
+}
+
+func TestApiRoutesAuthentication(t *testing.T) {
+	hub, user := beszelTests.GetHubWithUser(t)
+	defer hub.Cleanup()
+
+	userToken, err := user.NewAuthToken()
+	require.NoError(t, err, "Failed to create auth token")
+
+	// Create test user and get auth token
+	user2, err := beszelTests.CreateUser(hub, "testuser@example.com", "password123")
+	require.NoError(t, err, "Failed to create test user")
+	user2Token, err := user2.NewAuthToken()
+	require.NoError(t, err, "Failed to create user2 auth token")
+
+	adminUser, err := beszelTests.CreateUserWithRole(hub, "admin@example.com", "password123", "admin")
+	require.NoError(t, err, "Failed to create admin user")
+	adminUserToken, err := adminUser.NewAuthToken()
+
+	readOnlyUser, err := beszelTests.CreateUserWithRole(hub, "readonly@example.com", "password123", "readonly")
+	require.NoError(t, err, "Failed to create readonly user")
+	readOnlyUserToken, err := readOnlyUser.NewAuthToken()
+	require.NoError(t, err, "Failed to create readonly user auth token")
+
+	superuser, err := beszelTests.CreateSuperuser(hub, "superuser@example.com", "password123")
+	require.NoError(t, err, "Failed to create superuser")
+	superuserToken, err := superuser.NewAuthToken()
+	require.NoError(t, err, "Failed to create superuser auth token")
+
+	// Create test system
+	system, err := beszelTests.CreateRecord(hub, "systems", map[string]any{
+		"name":  "test-system",
+		"users": []string{user.Id},
+		"host":  "127.0.0.1",
+	})
+	require.NoError(t, err, "Failed to create test system")
+
+	testAppFactory := func(t testing.TB) *pbTests.TestApp {
+		return hub.TestApp
+	}
+
+	scenarios := []beszelTests.ApiScenario{
+		// Auth Protected Routes - Should require authentication
+		{
+			Name:            "GET /config-yaml - no auth should fail",
+			Method:          http.MethodGet,
+			URL:             "/api/beszel/config-yaml",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /config-yaml - with user auth should fail",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/config-yaml",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  403,
+			ExpectedContent: []string{"The authorized record is not allowed to perform this action."},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /config-yaml - with admin auth should succeed",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/config-yaml",
+			Headers: map[string]string{
+				"Authorization": adminUserToken,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"test-system"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:            "GET /heartbeat-status - no auth should fail",
+			Method:          http.MethodGet,
+			URL:             "/api/beszel/heartbeat-status",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /heartbeat-status - with user auth should fail",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/heartbeat-status",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  403,
+			ExpectedContent: []string{"The authorized record is not allowed to perform this action."},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /heartbeat-status - with admin auth should succeed",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/heartbeat-status",
+			Headers: map[string]string{
+				"Authorization": adminUserToken,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{`"enabled":false`},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "POST /test-heartbeat - with user auth should fail",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/test-heartbeat",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  403,
+			ExpectedContent: []string{"The authorized record is not allowed to perform this action."},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "POST /test-heartbeat - with admin auth should report disabled state",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/test-heartbeat",
+			Headers: map[string]string{
+				"Authorization": adminUserToken,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"Heartbeat not configured"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:            "GET /universal-token - no auth should fail",
+			Method:          http.MethodGet,
+			URL:             "/api/beszel/universal-token",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /universal-token - with auth should succeed",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/universal-token",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"active", "token", "permanent"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /universal-token - enable permanent should succeed",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/universal-token?enable=1&permanent=1&token=permanent-token-123",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"permanent\":true", "permanent-token-123"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /universal-token - superuser should fail",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/universal-token",
+			Headers: map[string]string{
+				"Authorization": superuserToken,
+			},
+			ExpectedStatus:  403,
+			ExpectedContent: []string{"Superusers cannot use universal tokens"},
+			TestAppFactory: func(t testing.TB) *pbTests.TestApp {
+				return hub.TestApp
+			},
+		},
+		{
+			Name:   "GET /universal-token - with readonly auth should fail",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/universal-token",
+			Headers: map[string]string{
+				"Authorization": readOnlyUserToken,
+			},
+			ExpectedStatus:  403,
+			ExpectedContent: []string{"The authorized record is not allowed to perform this action."},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "POST /smart/refresh - missing system should fail 400 with user auth",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/smart/refresh",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  400,
+			ExpectedContent: []string{"Invalid", "system", "parameter"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "POST /smart/refresh - with readonly auth should fail",
+			Method: http.MethodPost,
+			URL:    fmt.Sprintf("/api/beszel/smart/refresh?system=%s", system.Id),
+			Headers: map[string]string{
+				"Authorization": readOnlyUserToken,
+			},
+			ExpectedStatus:  403,
+			ExpectedContent: []string{"The authorized record is not allowed to perform this action."},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "POST /smart/refresh - non-user system should fail",
+			Method: http.MethodPost,
+			URL:    fmt.Sprintf("/api/beszel/smart/refresh?system=%s", system.Id),
+			Headers: map[string]string{
+				"Authorization": user2Token,
+			},
+			ExpectedStatus:  404,
+			ExpectedContent: []string{"The requested resource wasn't found."},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "POST /smart/refresh - good user should pass validation",
+			Method: http.MethodPost,
+			URL:    fmt.Sprintf("/api/beszel/smart/refresh?system=%s", system.Id),
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  500,
+			ExpectedContent: []string{"Something went wrong while processing your request."},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:            "POST /user-alerts - no auth should fail",
+			Method:          http.MethodPost,
+			URL:             "/api/beszel/user-alerts",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "CPU",
+				"value":   80,
+				"min":     10,
+				"systems": []string{system.Id},
+			}),
+		},
+		{
+			Name:   "POST /user-alerts - with auth should succeed",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "CPU",
+				"value":   80,
+				"min":     10,
+				"systems": []string{system.Id},
+			}),
+		},
+		{
+			Name:            "DELETE /user-alerts - no auth should fail",
+			Method:          http.MethodDelete,
+			URL:             "/api/beszel/user-alerts",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "CPU",
+				"systems": []string{system.Id},
+			}),
+		},
+		{
+			Name:   "DELETE /user-alerts - with auth should succeed",
+			Method: http.MethodDelete,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"success\":true"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "CPU",
+				"systems": []string{system.Id},
+			}),
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				// Create an alert to delete
+				beszelTests.CreateRecord(app, "alerts", map[string]any{
+					"name":   "CPU",
+					"system": system.Id,
+					"user":   user.Id,
+					"value":  80,
+					"min":    10,
+				})
+			},
+		},
+		{
+			Name:            "GET /containers/logs - no auth should fail",
+			Method:          http.MethodGet,
+			URL:             "/api/beszel/containers/logs?system=test-system&container=abababababab",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:            "GET /containers/logs - request for valid non-user system should fail",
+			Method:          http.MethodGet,
+			URL:             fmt.Sprintf("/api/beszel/containers/logs?system=%s&container=abababababab", system.Id),
+			ExpectedStatus:  404,
+			ExpectedContent: []string{"The requested resource wasn't found."},
+			TestAppFactory:  testAppFactory,
+			Headers: map[string]string{
+				"Authorization": user2Token,
+			},
+		},
+		{
+			Name:            "GET /containers/info - request for valid non-user system should fail",
+			Method:          http.MethodGet,
+			URL:             fmt.Sprintf("/api/beszel/containers/info?system=%s&container=abababababab", system.Id),
+			ExpectedStatus:  404,
+			ExpectedContent: []string{"The requested resource wasn't found."},
+			TestAppFactory:  testAppFactory,
+			Headers: map[string]string{
+				"Authorization": user2Token,
+			},
+		},
+		{
+			Name:            "GET /containers/info - SHARE_ALL_SYSTEMS allows non-member user",
+			Method:          http.MethodGet,
+			URL:             fmt.Sprintf("/api/beszel/containers/info?system=%s&container=abababababab", system.Id),
+			ExpectedStatus:  500,
+			ExpectedContent: []string{"Something went wrong while processing your request."},
+			TestAppFactory:  testAppFactory,
+			Headers: map[string]string{
+				"Authorization": user2Token,
+			},
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				t.Setenv("SHARE_ALL_SYSTEMS", "true")
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				t.Setenv("SHARE_ALL_SYSTEMS", "")
+			},
+		},
+		{
+			Name:   "GET /containers/logs - with auth but missing system param should fail",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/containers/logs?container=abababababab",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  400,
+			ExpectedContent: []string{"Invalid", "parameter"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /containers/logs - with auth but missing container param should fail",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/containers/logs?system=test-system",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  400,
+			ExpectedContent: []string{"Invalid", "parameter"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /containers/logs - with auth but invalid system should fail",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/containers/logs?system=invalid-system&container=0123456789ab",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  404,
+			ExpectedContent: []string{"The requested resource wasn't found."},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /containers/logs - traversal container should fail validation",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/containers/logs?system=" + system.Id + "&container=..%2F..%2Fversion",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  400,
+			ExpectedContent: []string{"Invalid", "parameter"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /containers/info - traversal container should fail validation",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/containers/info?system=" + system.Id + "&container=../../version?x=",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  400,
+			ExpectedContent: []string{"Invalid", "parameter"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /containers/info - non-hex container should fail validation",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/containers/info?system=" + system.Id + "&container=container_name",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  400,
+			ExpectedContent: []string{"Invalid", "parameter"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /containers/logs - good user should pass validation",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/containers/logs?system=" + system.Id + "&container=0123456789ab",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  500,
+			ExpectedContent: []string{"Something went wrong while processing your request."},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /containers/info - good user should pass validation",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/containers/info?system=" + system.Id + "&container=0123456789ab",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  500,
+			ExpectedContent: []string{"Something went wrong while processing your request."},
+			TestAppFactory:  testAppFactory,
+		},
+		// /systemd routes
+		{
+			Name:            "GET /systemd/info - no auth should fail",
+			Method:          http.MethodGet,
+			URL:             fmt.Sprintf("/api/beszel/systemd/info?system=%s&service=nginx.service", system.Id),
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:            "GET /systemd/info - request for valid non-user system should fail",
+			Method:          http.MethodGet,
+			URL:             fmt.Sprintf("/api/beszel/systemd/info?system=%s&service=nginx.service", system.Id),
+			ExpectedStatus:  404,
+			ExpectedContent: []string{"The requested resource wasn't found."},
+			TestAppFactory:  testAppFactory,
+			Headers: map[string]string{
+				"Authorization": user2Token,
+			},
+		},
+		{
+			Name:   "GET /systemd/info - with auth but missing system param should fail",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/systemd/info?service=nginx.service",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  400,
+			ExpectedContent: []string{"Invalid", "parameter"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /systemd/info - with auth but missing service param should fail",
+			Method: http.MethodGet,
+			URL:    fmt.Sprintf("/api/beszel/systemd/info?system=%s", system.Id),
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  400,
+			ExpectedContent: []string{"Invalid", "parameter"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /systemd/info - with auth but invalid system should fail",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/systemd/info?system=invalid-system&service=nginx.service",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  404,
+			ExpectedContent: []string{"The requested resource wasn't found."},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /systemd/info - service not in systemd_services collection should fail",
+			Method: http.MethodGet,
+			URL:    fmt.Sprintf("/api/beszel/systemd/info?system=%s&service=notregistered.service", system.Id),
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  404,
+			ExpectedContent: []string{"The requested resource wasn't found."},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /systemd/info - with auth and existing service record should pass validation",
+			Method: http.MethodGet,
+			URL:    fmt.Sprintf("/api/beszel/systemd/info?system=%s&service=nginx.service", system.Id),
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  500,
+			ExpectedContent: []string{"Something went wrong while processing your request."},
+			TestAppFactory:  testAppFactory,
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.CreateRecord(app, "systemd_services", map[string]any{
+					"system": system.Id,
+					"name":   "nginx.service",
+					"state":  0,
+					"sub":    1,
+				})
+			},
+		},
+
+		// Auth Optional Routes - Should work without authentication
+		{
+			Name:            "GET /getkey - no auth should fail",
+			Method:          http.MethodGet,
+			URL:             "/api/beszel/getkey",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /getkey - with auth should also succeed",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/getkey",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"key\":", "\"v\":"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /info - should return the same as /getkey",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/info",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"key\":", "\"v\":"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:            "GET /first-run - no auth should succeed",
+			Method:          http.MethodGet,
+			URL:             "/api/beszel/first-run",
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"firstRun\":false"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /first-run - with auth should also succeed",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/first-run",
+			Headers: map[string]string{
+				"Authorization": userToken,
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"firstRun\":false"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:            "GET /agent-connect - no auth should succeed (websocket upgrade fails but route is accessible)",
+			Method:          http.MethodGet,
+			URL:             "/api/beszel/agent-connect",
+			ExpectedStatus:  400,
+			ExpectedContent: []string{},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "POST /test-notification - invalid auth token should fail",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/test-notification",
+			Body: jsonReader(map[string]any{
+				"url": "generic://127.0.0.1",
+			}),
+			Headers: map[string]string{
+				"Authorization": "invalid-token",
+			},
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "POST /user-alerts - invalid auth token should fail",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/user-alerts",
+			Headers: map[string]string{
+				"Authorization": "invalid-token",
+			},
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+			Body: jsonReader(map[string]any{
+				"name":    "CPU",
+				"value":   80,
+				"min":     10,
+				"systems": []string{system.Id},
+			}),
+		},
+		// this works but diff behavior on prod vs dev.
+		// dev returns 502; prod returns 200 with static html page 404
+		// TODO: align dev and prod behavior and re-enable this test
+		// {
+		// 	Name:               "GET /update - shouldn't exist without CHECK_UPDATES env var",
+		// 	Method:             http.MethodGet,
+		// 	URL:                "/api/beszel/update",
+		// 	NotExpectedContent: []string{"v:", "\"v\":"},
+		// 	ExpectedStatus: 502,
+		// 	TestAppFactory: testAppFactory,
+		// },
+	}
+
+	for _, scenario := range scenarios {
+		scenario.Test(t)
+	}
+}
+
+func TestFirstUserCreation(t *testing.T) {
+	t.Run("CreateUserEndpoint available when no users exist", func(t *testing.T) {
+		hub, _ := beszelTests.NewTestHub(t.TempDir())
+		defer hub.Cleanup()
+
+		hub.StartHub()
+
+		testAppFactoryExisting := func(t testing.TB) *pbTests.TestApp {
+			return hub.TestApp
+		}
+
+		scenarios := []beszelTests.ApiScenario{
+			{
+				Name:   "POST /create-user - should be available when no users exist",
+				Method: http.MethodPost,
+				URL:    "/api/beszel/create-user",
+				Body: jsonReader(map[string]any{
+					"email":    "firstuser@example.com",
+					"password": "password123",
+				}),
+				ExpectedStatus:  200,
+				ExpectedContent: []string{"User created"},
+				TestAppFactory:  testAppFactoryExisting,
+				BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+					userCount, err := hub.CountRecords("users")
+					require.NoError(t, err)
+					require.Zero(t, userCount, "Should start with no users")
+					superusers, err := hub.FindAllRecords(core.CollectionNameSuperusers)
+					require.NoError(t, err)
+					require.EqualValues(t, 1, len(superusers), "Should start with one temporary superuser")
+					require.EqualValues(t, migrations.TempAdminEmail, superusers[0].GetString("email"), "Should have created one temporary superuser")
+				},
+				AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+					userCount, err := hub.CountRecords("users")
+					require.NoError(t, err)
+					require.EqualValues(t, 1, userCount, "Should have created one user")
+					superusers, err := hub.FindAllRecords(core.CollectionNameSuperusers)
+					require.NoError(t, err)
+					require.EqualValues(t, 1, len(superusers), "Should have created one superuser")
+					require.EqualValues(t, "firstuser@example.com", superusers[0].GetString("email"), "Should have created one superuser")
+				},
+			},
+			{
+				Name:   "POST /create-user - should not be available when users exist",
+				Method: http.MethodPost,
+				URL:    "/api/beszel/create-user",
+				Body: jsonReader(map[string]any{
+					"email":    "firstuser@example.com",
+					"password": "password123",
+				}),
+				ExpectedStatus:  404,
+				ExpectedContent: []string{"wasn't found"},
+				TestAppFactory:  testAppFactoryExisting,
+			},
+		}
+
+		for _, scenario := range scenarios {
+			scenario.Test(t)
+		}
+	})
+
+	t.Run("CreateUserEndpoint not available when USER_EMAIL, USER_PASSWORD are set", func(t *testing.T) {
+		t.Setenv("BESZEL_HUB_USER_EMAIL", "me@example.com")
+		t.Setenv("BESZEL_HUB_USER_PASSWORD", "password123")
+
+		hub, _ := beszelTests.NewTestHub(t.TempDir())
+		defer hub.Cleanup()
+
+		hub.StartHub()
+
+		testAppFactory := func(t testing.TB) *pbTests.TestApp {
+			return hub.TestApp
+		}
+
+		scenario := beszelTests.ApiScenario{
+			Name:            "POST /create-user - should not be available when USER_EMAIL, USER_PASSWORD are set",
+			Method:          http.MethodPost,
+			URL:             "/api/beszel/create-user",
+			ExpectedStatus:  404,
+			ExpectedContent: []string{"wasn't found"},
+			TestAppFactory:  testAppFactory,
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				users, err := hub.FindAllRecords("users")
+				require.NoError(t, err)
+				require.EqualValues(t, 1, len(users), "Should start with one user")
+				require.EqualValues(t, "me@example.com", users[0].GetString("email"), "Should have created one user")
+				superusers, err := hub.FindAllRecords(core.CollectionNameSuperusers)
+				require.NoError(t, err)
+				require.EqualValues(t, 1, len(superusers), "Should start with one superuser")
+				require.EqualValues(t, "me@example.com", superusers[0].GetString("email"), "Should have created one superuser")
+			},
+			AfterTestFunc: func(t testing.TB, app *pbTests.TestApp, res *http.Response) {
+				users, err := hub.FindAllRecords("users")
+				require.NoError(t, err)
+				require.EqualValues(t, 1, len(users), "Should still have one user")
+				require.EqualValues(t, "me@example.com", users[0].GetString("email"), "Should have created one user")
+				superusers, err := hub.FindAllRecords(core.CollectionNameSuperusers)
+				require.NoError(t, err)
+				require.EqualValues(t, 1, len(superusers), "Should still have one superuser")
+				require.EqualValues(t, "me@example.com", superusers[0].GetString("email"), "Should have created one superuser")
+			},
+		}
+
+		scenario.Test(t)
+	})
+}
+
+func TestCreateUserEndpointAvailability(t *testing.T) {
+	t.Run("CreateUserEndpoint available when no users exist", func(t *testing.T) {
+		hub, _ := beszelTests.NewTestHub(t.TempDir())
+		defer hub.Cleanup()
+
+		// Ensure no users exist
+		userCount, err := hub.CountRecords("users")
+		require.NoError(t, err)
+		require.Zero(t, userCount, "Should start with no users")
+
+		hub.StartHub()
+
+		testAppFactory := func(t testing.TB) *pbTests.TestApp {
+			return hub.TestApp
+		}
+
+		scenario := beszelTests.ApiScenario{
+			Name:   "POST /create-user - should be available when no users exist",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/create-user",
+			Body: jsonReader(map[string]any{
+				"email":    "firstuser@example.com",
+				"password": "password123",
+			}),
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"User created"},
+			TestAppFactory:  testAppFactory,
+		}
+
+		scenario.Test(t)
+
+		// Verify user was created
+		userCount, err = hub.CountRecords("users")
+		require.NoError(t, err)
+		require.EqualValues(t, 1, userCount, "Should have created one user")
+	})
+
+	t.Run("CreateUserEndpoint not available when users exist", func(t *testing.T) {
+		hub, _ := beszelTests.NewTestHub(t.TempDir())
+		defer hub.Cleanup()
+
+		// Create a user first
+		_, err := beszelTests.CreateUser(hub, "existing@example.com", "password")
+		require.NoError(t, err)
+
+		hub.StartHub()
+
+		testAppFactory := func(t testing.TB) *pbTests.TestApp {
+			return hub.TestApp
+		}
+
+		scenario := beszelTests.ApiScenario{
+			Name:   "POST /create-user - should not be available when users exist",
+			Method: http.MethodPost,
+			URL:    "/api/beszel/create-user",
+			Body: jsonReader(map[string]any{
+				"email":    "another@example.com",
+				"password": "password123",
+			}),
+			ExpectedStatus:  404,
+			ExpectedContent: []string{"wasn't found"},
+			TestAppFactory:  testAppFactory,
+		}
+
+		scenario.Test(t)
+	})
+}
+
+func TestAutoLoginMiddleware(t *testing.T) {
+	var hubs []*beszelTests.TestHub
+
+	defer func() {
+		for _, hub := range hubs {
+			hub.Cleanup()
+		}
+	}()
+
+	t.Setenv("AUTO_LOGIN", "user@test.com")
+
+	testAppFactory := func(t testing.TB) *pbTests.TestApp {
+		hub, _ := beszelTests.NewTestHub(t.TempDir())
+		hubs = append(hubs, hub)
+		hub.StartHub()
+		return hub.TestApp
+	}
+
+	scenarios := []beszelTests.ApiScenario{
+		{
+			Name:            "GET /getkey - without auto login should fail",
+			Method:          http.MethodGet,
+			URL:             "/api/beszel/getkey",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:            "GET /getkey - with auto login should fail if no matching user",
+			Method:          http.MethodGet,
+			URL:             "/api/beszel/getkey",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:            "GET /getkey - with auto login should succeed",
+			Method:          http.MethodGet,
+			URL:             "/api/beszel/getkey",
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"key\":", "\"v\":"},
+			TestAppFactory:  testAppFactory,
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.CreateUser(app, "user@test.com", "password123")
+			},
+		},
+	}
+
+	for _, scenario := range scenarios {
+		scenario.Test(t)
+	}
+}
+
+func TestTrustedHeaderMiddleware(t *testing.T) {
+	var hubs []*beszelTests.TestHub
+
+	defer func() {
+		for _, hub := range hubs {
+			hub.Cleanup()
+		}
+	}()
+
+	t.Setenv("TRUSTED_AUTH_HEADER", "X-Beszel-Trusted")
+
+	testAppFactory := func(t testing.TB) *pbTests.TestApp {
+		hub, _ := beszelTests.NewTestHub(t.TempDir())
+		hubs = append(hubs, hub)
+		hub.StartHub()
+		return hub.TestApp
+	}
+
+	scenarios := []beszelTests.ApiScenario{
+		{
+			Name:            "GET /getkey - without trusted header should fail",
+			Method:          http.MethodGet,
+			URL:             "/api/beszel/getkey",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /getkey - with trusted header should fail if no matching user",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/getkey",
+			Headers: map[string]string{
+				"X-Beszel-Trusted": "user@test.com",
+			},
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+		},
+		{
+			Name:   "GET /getkey - with trusted header should succeed",
+			Method: http.MethodGet,
+			URL:    "/api/beszel/getkey",
+			Headers: map[string]string{
+				"X-Beszel-Trusted": "user@test.com",
+			},
+			ExpectedStatus:  200,
+			ExpectedContent: []string{"\"key\":", "\"v\":"},
+			TestAppFactory:  testAppFactory,
+			BeforeTestFunc: func(t testing.TB, app *pbTests.TestApp, e *core.ServeEvent) {
+				beszelTests.CreateUser(app, "user@test.com", "password123")
+			},
+		},
+	}
+
+	for _, scenario := range scenarios {
+		scenario.Test(t)
+	}
+}
+
+func TestUpdateEndpoint(t *testing.T) {
+	t.Setenv("CHECK_UPDATES", "true")
+
+	hub, _ := beszelTests.NewTestHub(t.TempDir())
+	defer hub.Cleanup()
+	hub.StartHub()
+
+	// Create test user and get auth token
+	// user, err := beszelTests.CreateUser(hub, "testuser@example.com", "password123")
+	// require.NoError(t, err, "Failed to create test user")
+	// userToken, err := user.NewAuthToken()
+
+	testAppFactory := func(t testing.TB) *pbTests.TestApp {
+		return hub.TestApp
+	}
+
+	scenarios := []beszelTests.ApiScenario{
+		{
+			Name:            "update endpoint shouldn't work without auth",
+			Method:          http.MethodGet,
+			URL:             "/api/beszel/update",
+			ExpectedStatus:  401,
+			ExpectedContent: []string{"requires valid"},
+			TestAppFactory:  testAppFactory,
+		},
+		// leave this out for now since it actually makes a request to github
+		// {
+		// 	Name:   "GET /update - with valid auth should succeed",
+		// 	Method: http.MethodGet,
+		// 	URL:    "/api/beszel/update",
+		// 	Headers: map[string]string{
+		// 		"Authorization": userToken,
+		// 	},
+		// 	ExpectedStatus:  200,
+		// 	ExpectedContent: []string{`"v":`},
+		// 	TestAppFactory:  testAppFactory,
+		// },
+	}
+
+	for _, scenario := range scenarios {
+		scenario.Test(t)
+	}
+}