🦞 OpenClaw Dashboard

See your agent think. The Grafana for your personal AI agent.

Real-time observability dashboard for OpenClaw/Moltbot AI agents. One file. Zero config. Just run it.

---

⚡ Quick Start

pip install openclaw-dashboard
openclaw-dashboard

Or run directly:

pip install flask
python3 dashboard.py

Opens at http://localhost:8900 — auto-detects your OpenClaw workspace.

---

✨ Features

Tab	What it shows
🌊 Flow	Real-time animated SVG showing data flow: You → Channels → Gateway → Brain → Tools → Infrastructure
Overview	Model, sessions, crons, tokens, memory, ❤️ health checks (auto-refresh via SSE), 🔥 activity heatmap (GitHub-style), recent logs
📊 Usage	Token/cost tracking — bar chart of tokens per day (14 days), today/week/month totals, cost breakdown. With OTLP: real token counts, actual cost, avg run duration, messages processed, model breakdown
Sessions	All active agent sessions with model, channel, token usage, last activity
Crons	Scheduled jobs with status, schedule, last run, next run, duration
Logs	Parsed JSON logs with color-coded levels, configurable line count, real-time SSE streaming
Memory	Clickable file browser for SOUL.md, MEMORY.md, AGENTS.md, daily memory files
📜 Transcripts	Session transcript viewer — browse .jsonl files, click to see chat-bubble conversation view with color-coded roles, expand/collapse

Flow Visualization

The Flow tab is the star — a live animated architecture diagram that lights up in real-time as your agent processes messages:

• 🟣 Purple particles — your message entering through a channel
• 🔵 Blue particles — request flowing to the brain
• 🟡 Yellow particles — tool calls (exec, browser, search, cron, tts, memory)
• 🟢 Green particles — response flowing back to you
• 🔴 Red flash — errors
• 🔵 Cyan pulses — infrastructure layer activity (network, storage, runtime)

New in v0.2: OTLP Receiver + Full Observability

• 📡 OTLP Receiver — Dashboard becomes a lightweight OTel collector. Point OpenClaw at it, get real metrics. No Grafana/Prometheus needed.
• 🔥 Activity Heatmap — GitHub-style 7×24 grid showing when your agent is busiest. Pure CSS, no libraries.
• ❤️ Health Checks — Gateway, disk, memory, uptime, OTLP status at a glance. Auto-refreshes every 30s via SSE.
• 📊 Real Token/Cost Tracking — With OTLP: real token counts, actual cost, model breakdown, avg run duration.
• 📜 Transcript Viewer — Read your agent's conversations in a beautiful chat-bubble UI. Color-coded roles, expand/collapse for long messages.

---

🤔 What Makes This Different?

	OpenClaw Dashboard	Langfuse	AgentOps
Install	pip install openclaw-dashboard	Docker + Postgres	SDK + cloud account
Config	Zero. Auto-detects everything.	Database URLs, API keys	API keys, SDK init
Focus	Personal AI agent	Enterprise LLM apps	Enterprise agent monitoring
Memory-first	✅ Browse SOUL.md, MEMORY.md, daily notes	❌	❌
Single file	✅ One Python file, one dependency	❌ Multi-service	❌ Cloud service
Transcripts	✅ Chat-bubble viewer built-in	✅ (needs SDK)	✅ (needs SDK)
Cost tracking	✅ Zero config (OTLP or log parsing)	✅ (needs SDK)	✅ (needs SDK)
Built-in OTel collector	✅ OTLP/HTTP receiver	❌	❌

TL;DR: Langfuse and AgentOps are great for teams building LLM products. OpenClaw Dashboard is for the person running a personal AI agent on their own machine — zero instrumentation, zero config, memory-first. It's the Grafana for your personal AI agent.

---

📡 Real-time Metrics (OpenTelemetry)

The dashboard can act as a lightweight OpenTelemetry collector — no need for Grafana, Prometheus, or a separate OTel Collector. Just point OpenClaw at the dashboard.

Setup

1. Install OTLP support:

pip install openclaw-dashboard[otel]

2. Configure OpenClaw — add one line to your config:

diagnostics:
  otel:
    endpoint: http://localhost:8900

That's it! The dashboard now receives real-time metrics directly from OpenClaw.

What you get

Metric	Source	What it shows
Token counts per day	openclaw.tokens	Real input/output/total token usage (bar chart)
Cost per day	openclaw.cost.usd	Actual cost from your provider
Avg run duration	openclaw.run.duration_ms	How long model completions take
Messages processed	openclaw.message.processed	Message throughput
Model breakdown	attributes	Which models are being used and how much
OTLP Connected indicator	health check	Green when data is flowing

OTLP Endpoints

• POST /v1/metrics — receives OTLP/HTTP protobuf metric data
• POST /v1/traces — receives OTLP/HTTP protobuf trace data

Without OTLP

Everything still works! The dashboard falls back to parsing session JSONL files for token estimates. OTLP just gives you real numbers instead of estimates.

Persistence

Metrics are stored in-memory (capped at ~10K entries per category, 14-day retention) and auto-persisted to {workspace}/.openclaw-dashboard-metrics.json every 60 seconds. Override the path with --metrics-file or OPENCLAW_METRICS_FILE.

---

⚙️ Configuration

CLI Arguments

openclaw-dashboard --port 9000          # Custom port (default: 8900)
openclaw-dashboard --host 127.0.0.1     # Bind to localhost only
openclaw-dashboard --workspace ~/mybot  # Custom workspace path
openclaw-dashboard --log-dir /var/log   # Custom log directory
openclaw-dashboard --sessions-dir ~/data # Custom sessions directory
openclaw-dashboard --metrics-file ~/m.json # Custom metrics persistence path
openclaw-dashboard --name "Alice"       # Your name in Flow visualization

Environment Variables

Variable	Description	Default
OPENCLAW_HOME	Agent workspace directory	Auto-detected
OPENCLAW_WORKSPACE	Alternative to OPENCLAW_HOME	Auto-detected
OPENCLAW_SESSIONS_DIR	Sessions directory (.jsonl transcripts)	Auto-detected
OPENCLAW_LOG_DIR	Log directory	/tmp/moltbot
OPENCLAW_METRICS_FILE	Metrics persistence file path	{workspace}/.openclaw-dashboard-metrics.json
OPENCLAW_USER	Your name in Flow tab	You

Auto-Detection

If no paths are configured, the dashboard automatically searches for:

1. Workspace: Checks ~/.clawdbot/agents/main/config.json → ~/.clawdbot/workspace → ~/clawd → ~/openclaw → current directory. Looks for SOUL.md, AGENTS.md, MEMORY.md, or memory/ directory.
2. Logs: Checks /tmp/moltbot → /tmp/openclaw → ~/.clawdbot/logs
3. Sessions: Reads from ~/.clawdbot/agents/main/sessions/
4. Crons: Reads from ~/.clawdbot/cron/jobs.json

---

🏗️ How It Works

The dashboard is a single-file Flask app that reads directly from your OpenClaw/Moltbot data directories:

Your Agent (Moltbot)          OpenClaw Dashboard
┌─────────────────┐          ┌─────────────────┐
│ Writes logs to   │─────────▶│ Reads & parses   │
│ /tmp/moltbot/    │          │ JSON log lines   │
│                  │          │                  │
│ Stores sessions  │─────────▶│ Lists sessions   │
│ in ~/.clawdbot/  │          │ with metadata    │
│                  │          │                  │
│ Saves crons to   │─────────▶│ Shows schedules  │
│ cron/jobs.json   │          │ and status       │
│                  │          │                  │
│ Agent workspace  │─────────▶│ Browses memory   │
│ SOUL.md, etc.    │          │ files inline     │
└─────────────────┘          └─────────────────┘

Real-time streaming uses tail -f piped through Server-Sent Events (SSE) — no WebSockets, no dependencies, just works.

---

📦 Installation Options

pip (recommended)

pip install openclaw-dashboard
openclaw-dashboard

From source

git clone https://github.com/vivekchand/openclaw-dashboard.git
cd openclaw-dashboard
pip install -r requirements.txt
python3 dashboard.py

One-liner

curl -sSL https://raw.githubusercontent.com/vivekchand/openclaw-dashboard/main/install.sh | bash

---

🔧 Requirements

• Python 3.8+
• Flask (only required dependency)
• opentelemetry-proto + protobuf (optional, for OTLP receiver — pip install openclaw-dashboard[otel])
• OpenClaw/Moltbot running on the same machine (reads its logs and state files)
• Linux/macOS (uses tail, df, free, /proc/loadavg)

---

📄 License

MIT — do whatever you want with it.

---

🙏 Credits

• Built by Vivek Chand as part of the OpenClaw ecosystem
• Powered by OpenClaw/Moltbot
• The Flow visualization was inspired by watching an AI agent actually think

---

🦞 See your agent think
_{Star this repo if you find it useful!}