feat: access-log telemetry, GPU metrics, filterable panels, trend charts

- Parse Ollama server.log for API latency, request counts and endpoint activity
- Add GPU telemetry via agputop JSON on macOS; improve nvidia-smi parsing on Linux
- Expose token throughput, request latency and active model insights in overview panel
- Add log path and request timeout config options (OLLAMON_LOG_PATH, OLLAMON_REQUEST_TIMEOUT_MS)
- Update README with screenshot placeholder, full config reference, GPU and log docs

Author: hbasria

README.md

@@ -1,27 +1,117 @@
 # ollamon
 
-Terminal monitor for Ollama nodes.
+`ollamon` is a terminal monitor for Ollama nodes.
 
-## install (one-line)
+It provides a TUI focused on operational visibility:
 
+- installed models
+- running models
+- CPU, memory, disk, and GPU activity
+- Ollama access-log telemetry
+- lightweight operational insights
+
+## Screenshot
+
+Add your screenshot file and update the path below if needed:
+
+![ollamon screenshot](./docs/screenshot.png)
+
+## Features
+
+- Live overview for host and Ollama runtime state
+- Running and installed model panels
+- Access-log telemetry from Ollama `server.log`
+- Filterable model views
+- Mini trend charts for CPU, memory, disk, and GPU
+- GPU metrics on macOS via [`agputop`](https://github.com/hbasria/agputop)
+
+## Install
+
+One-line install:
+
+```bash
 curl -fsSL https://raw.githubusercontent.com/hbasria/ollamon/main/scripts/install.sh | sh
+```
 
-Sadece kurulum (çalıştırmadan):
+Install only, without launching:
 
+```bash
 curl -fsSL https://raw.githubusercontent.com/hbasria/ollamon/main/scripts/install.sh | sh -s -- --no-run
+```
 
-Sürüm vererek kurulum:
+Install a specific version:
 
+```bash
 curl -fsSL https://raw.githubusercontent.com/hbasria/ollamon/main/scripts/install.sh | sh -s -- --version v0.1.0
+```
 
 Supported targets:
 
-- darwin/amd64
-- darwin/arm64 (Apple Silicon)
-- linux/amd64
-- linux/arm64
+- `darwin/amd64`
+- `darwin/arm64`
+- `linux/amd64`
+- `linux/arm64`
+
+## Run
+
+```bash
+make build
+./bin/ollamon
+```
+
+For development:
+
+```bash
+make run
+```
+
+## Configuration
+
+Environment variables:
+
+- `OLLAMA_HOST`
+- `OLLAMON_INTERVAL_MS`
+- `OLLAMON_REQUEST_TIMEOUT_MS`
+- `OLLAMON_DISK_PATH`
+- `OLLAMON_LOG_PATH`
+- `OLLAMON_COMPACT`
+
+Example:
+
+```bash
+OLLAMA_HOST=http://127.0.0.1:11434 \
+OLLAMON_INTERVAL_MS=2000 \
+OLLAMON_REQUEST_TIMEOUT_MS=5000 \
+make run
+```
+
+## Logs
+
+By default, `ollamon` reads Ollama access logs from:
+
+```bash
+~/.ollama/logs/server.log
+```
+
+You can override this with:
+
+```bash
+OLLAMON_LOG_PATH=/path/to/server.log
+```
+
+## GPU Metrics
+
+On macOS, GPU telemetry is collected from [`agputop`](https://github.com/hbasria/agputop) using its JSON output.
+
+Example command:
+
+```bash
+agputop --json
+```
+
+If `agputop` is not available, `ollamon` falls back to basic GPU detection.
 
-## run
+## Notes
 
-go mod tidy
-go run ./cmd/ollamon
+- Access-log-derived telemetry can show API latency, request counts, and endpoint activity.
+- Token throughput is only shown when the underlying telemetry source exposes it.

docs/screenshot.png

@@ -36,17 +36,19 @@ type Model struct {
 	cpuHistory  []float64
 	memHistory  []float64
 	diskHistory []float64
+	gpuHistory  []float64
 }
 
 func New(cfg config.Config) Model {
 	return Model{
 		cfg:         cfg,
-		client:      ollama.NewClient(cfg.BaseURL),
+		client:      ollama.NewClientWithTimeout(cfg.BaseURL, cfg.RequestTimeout),
 		width:       120,
 		height:      40,
 		cpuHistory:  make([]float64, 0, 40),
 		memHistory:  make([]float64, 0, 40),
 		diskHistory: make([]float64, 0, 40),
+		gpuHistory:  make([]float64, 0, 40),
 	}
 }
 
@@ -97,6 +99,7 @@ func (m Model) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 		m.cpuHistory = pushHistory(m.cpuHistory, msg.sample.Host.CPUPercent, 48)
 		m.memHistory = pushHistory(m.memHistory, msg.sample.Host.MemoryUsedPercent, 48)
 		m.diskHistory = pushHistory(m.diskHistory, msg.sample.Host.DiskUsedPercent, 48)
+		m.gpuHistory = pushHistory(m.gpuHistory, msg.sample.Host.GPU.UtilizationPct, 48)
 	case tickMsg:
 		return m, tea.Batch(m.refreshCmd(), tickCmd(m.cfg.Interval))
 	}
@@ -197,6 +200,7 @@ func (m Model) renderOverview() string {
 	cpuBar := util.PercentBar(s.Host.CPUPercent, 18, "█", "░")
 	memBar := util.PercentBar(s.Host.MemoryUsedPercent, 18, "█", "░")
 	diskBar := util.PercentBar(s.Host.DiskUsedPercent, 18, "█", "░")
+	gpuBar := util.PercentBar(s.Host.GPU.UtilizationPct, 18, "█", "░")
 
 	lines := []string{
 		tui.Title.Render("Overview"),
@@ -208,9 +212,11 @@ func (m Model) renderOverview() string {
 		fmt.Sprintf("%s %5.1f%% %s", tui.MetricLabel.Render("CPU"), s.Host.CPUPercent, colorizePercent(s.Host.CPUPercent, cpuBar)),
 		fmt.Sprintf("%s %5.1f%% %s", tui.MetricLabel.Render("MEM"), s.Host.MemoryUsedPercent, colorizePercent(s.Host.MemoryUsedPercent, memBar)),
 		fmt.Sprintf("%s %5.1f%% %s", tui.MetricLabel.Render("DSK"), s.Host.DiskUsedPercent, colorizePercent(s.Host.DiskUsedPercent, diskBar)),
+		fmt.Sprintf("%s %5.1f%% %s", tui.MetricLabel.Render("GPU"), s.Host.GPU.UtilizationPct, colorizePercent(s.Host.GPU.UtilizationPct, gpuBar)),
 		"",
 		fmt.Sprintf("%s %.1f / %.1f GB", tui.MetricLabel.Render("Memory"), s.Host.MemoryUsedGB, s.Host.MemoryTotalGB),
 		fmt.Sprintf("%s %.1f / %.1f GB", tui.MetricLabel.Render("Disk"), s.Host.DiskUsedGB, s.Host.DiskTotalGB),
+		fmt.Sprintf("%s %.1fW @ %.0fMHz", tui.MetricLabel.Render("GPU Pwr"), s.Host.GPU.PowerW, s.Host.GPU.FrequencyMHz),
 		fmt.Sprintf("%s %.2f / %.2f", tui.MetricLabel.Render("Load"), s.Host.Load1, s.Host.Load5),
 		fmt.Sprintf("%s %d installed / %d running", tui.MetricLabel.Render("Models"), len(s.Installed), len(s.Running)),
 	}
@@ -229,12 +235,14 @@ func (m Model) renderTelemetry() string {
 	cpuSpark := util.Sparkline(m.cpuHistory, 32)
 	memSpark := util.Sparkline(m.memHistory, 32)
 	diskSpark := util.Sparkline(m.diskHistory, 32)
+	gpuSpark := util.Sparkline(m.gpuHistory, 32)
 
 	lines := []string{
 		tui.Title.Render("Trends"),
 		fmt.Sprintf("%s %s", tui.MetricLabel.Render("CPU"), tui.Accent.Render(cpuSpark)),
 		fmt.Sprintf("%s %s", tui.MetricLabel.Render("MEM"), tui.Accent.Render(memSpark)),
 		fmt.Sprintf("%s %s", tui.MetricLabel.Render("DSK"), tui.Accent.Render(diskSpark)),
+		fmt.Sprintf("%s %s", tui.MetricLabel.Render("GPU"), tui.Accent.Render(gpuSpark)),
 		"",
 		tui.Title.Render("Telemetry"),
 		fmt.Sprintf("%s %s", tui.MetricLabel.Render("Token/s"), renderTokenRate(s)),
@@ -254,9 +262,9 @@ func (m Model) renderTelemetry() string {
 	}
 
 	if s.Time.IsZero() {
-		lines = append(lines, "", tui.Dim.Render("İlk örnek bekleniyor..."))
+		lines = append(lines, "", tui.Dim.Render("Waiting for first sample..."))
 	} else if s.LogStats.RequestCount == 0 {
-		lines = append(lines, "", tui.Dim.Render("Log erişim kaydı bekleniyor."))
+		lines = append(lines, "", tui.Dim.Render("Waiting for access log entries."))
 	}
 
 	return tui.Box.Width(rightW).Render(strings.Join(lines, "\n"))
@@ -273,15 +281,15 @@ func (m Model) renderInsights() string {
 	insights := make([]string, 0, 6)
 
 	if s.Host.MemoryUsedPercent > 85 {
-		insights = append(insights, "• RAM kullanımı yüksek. Büyük model yüklemelerinde yavaşlama olabilir.")
+		insights = append(insights, "• RAM usage is high. Large model loads may slow down.")
 	} else if s.Host.MemoryUsedPercent > 70 {
-		insights = append(insights, "• RAM kullanımı orta-yüksek. Keep-alive süreleri gözden geçirilebilir.")
+		insights = append(insights, "• RAM usage is moderately high. Keep-alive settings may need review.")
 	} else {
-		insights = append(insights, "• RAM tarafı şu an rahat görünüyor.")
+		insights = append(insights, "• RAM usage is currently in a comfortable range.")
 	}
 
 	if s.Host.DiskUsedPercent > 80 {
-		insights = append(insights, "• Disk doluluğu kritik eşiğe yaklaşıyor. Eski modeller temizlenebilir.")
+		insights = append(insights, "• Disk usage is approaching a critical threshold. Old models may need cleanup.")
 	}
 
 	if len(s.Installed) > 0 {
@@ -291,25 +299,25 @@ func (m Model) renderInsights() string {
 				idx = i
 			}
 		}
-		insights = append(insights, fmt.Sprintf("• En büyük model: %s (%s)", s.Installed[idx].Name, util.BytesToHuman(s.Installed[idx].Size)))
+		insights = append(insights, fmt.Sprintf("• Largest installed model: %s (%s)", s.Installed[idx].Name, util.BytesToHuman(s.Installed[idx].Size)))
 	}
 
 	if len(s.Running) > 0 {
 		rm := s.Running[0]
-		insights = append(insights, fmt.Sprintf("• Aktif örnek: %s, VRAM/RAM yükü yaklaşık %s", rm.Name, util.BytesToHuman(rm.SizeVRAM)))
+		insights = append(insights, fmt.Sprintf("• Active runtime: %s, estimated VRAM/RAM load %s", rm.Name, util.BytesToHuman(rm.SizeVRAM)))
 	} else {
-		insights = append(insights, "• Şu an bellekte aktif model görünmüyor.")
+		insights = append(insights, "• No active in-memory model is visible right now.")
 	}
 
 	if s.LogStats.RequestCount > 0 {
-		insights = append(insights, fmt.Sprintf("• Son %d log kaydında ortalama latency %s.", s.LogStats.RequestCount, renderLatency(s.LogStats.AvgLatency)))
+		insights = append(insights, fmt.Sprintf("• Average latency across the last %d log entries is %s.", s.LogStats.RequestCount, renderLatency(s.LogStats.AvgLatency)))
 		if len(s.LogStats.TopEndpoints) > 0 {
-			insights = append(insights, fmt.Sprintf("• En yoğun endpoint: %s (%d istek).", s.LogStats.TopEndpoints[0].Endpoint, s.LogStats.TopEndpoints[0].Count))
+			insights = append(insights, fmt.Sprintf("• Busiest endpoint: %s (%d requests).", s.LogStats.TopEndpoints[0].Endpoint, s.LogStats.TopEndpoints[0].Count))
 		}
 	}
 
 	if m.filter != "" {
-		insights = append(insights, fmt.Sprintf("• Filtre etkin: %q", m.filter))
+		insights = append(insights, fmt.Sprintf("• Active filter: %q", m.filter))
 	}
 
 	return strings.Join(insights, "\n")
@@ -325,7 +333,7 @@ func (m Model) renderRunning() string {
 	running = filterRunning(running, m.filter)
 
 	if len(running) == 0 {
-		lines = append(lines, tui.Dim.Render("Aktif model yok veya filtre eşleşmedi."))
+		lines = append(lines, tui.Dim.Render("No active model found or the filter did not match."))
 		return tui.Box.Width(leftW).Render(strings.Join(lines, "\n"))
 	}
 
@@ -359,7 +367,7 @@ func (m Model) renderInstalled() string {
 	models = filterInstalled(models, m.filter)
 
 	if len(models) == 0 {
-		lines = append(lines, tui.Dim.Render("Yüklü model bulunamadı veya filtre eşleşmedi."))
+		lines = append(lines, tui.Dim.Render("No installed model found or the filter did not match."))
 		return tui.Box.Width(rightW).Render(strings.Join(lines, "\n"))
 	}
 
@@ -394,7 +402,7 @@ func (m Model) renderLogPanel() string {
 	case m.last.LogError != "":
 		lines = append(lines, tui.Warn.Render(m.last.LogError))
 	case len(m.last.LogLines) == 0:
-		lines = append(lines, tui.Dim.Render("Log bulunamadı. OLLAMON_LOG_PATH ile özel yol verebilirsin."))
+		lines = append(lines, tui.Dim.Render("Log file not found. Use OLLAMON_LOG_PATH to provide a custom path."))
 	default:
 		logLines := m.last.LogLines
 		if len(logLines) > 10 {

internal/app/model.go

@@ -3,6 +3,7 @@ package app
 import (
 	"context"
 	"fmt"
+	"sync"
 	"time"
 
 	"github.com/example/ollamon/internal/config"
@@ -26,8 +27,6 @@ type Sample struct {
 
 func CollectSample(cfg config.Config, client *ollama.Client) (Sample, error) {
 	s := Sample{Time: time.Now(), BaseURL: client.BaseURL}
-	ctx, cancel := context.WithTimeout(context.Background(), 1500*time.Millisecond)
-	defer cancel()
 
 	hs, _ := system.Collect(cfg.RootDiskPath)
 	s.Host = hs
@@ -40,8 +39,36 @@ func CollectSample(cfg config.Config, client *ollama.Client) (Sample, error) {
 		s.LogError = logErr.Error()
 	}
 
-	installed, errTags := client.Tags(ctx)
-	running, errPS := client.PS(ctx)
+	timeout := cfg.RequestTimeout
+	if timeout <= 0 {
+		timeout = 5 * time.Second
+	}
+
+	var (
+		installed []ollama.Model
+		running   []ollama.RunningModel
+		errTags   error
+		errPS     error
+		wg        sync.WaitGroup
+	)
+
+	wg.Add(2)
+
+	go func() {
+		defer wg.Done()
+		ctx, cancel := context.WithTimeout(context.Background(), timeout)
+		defer cancel()
+		installed, errTags = client.Tags(ctx)
+	}()
+
+	go func() {
+		defer wg.Done()
+		ctx, cancel := context.WithTimeout(context.Background(), timeout)
+		defer cancel()
+		running, errPS = client.PS(ctx)
+	}()
+
+	wg.Wait()
 
 	s.Installed = installed
 	s.Running = running
@@ -52,7 +79,7 @@ func CollectSample(cfg config.Config, client *ollama.Client) (Sample, error) {
 		return s, nil
 	case errTags != nil && errPS != nil:
 		s.HealthError = fmt.Sprintf("tags: %v | ps: %v", errTags, errPS)
-		return s, fmt.Errorf("ollama erişilemedi: %s", s.HealthError)
+		return s, fmt.Errorf("ollama unavailable: %s", s.HealthError)
 	case errTags != nil:
 		s.HealthError = errTags.Error()
 		return s, nil