A beautiful, zero-dependency command center for OpenClaw AI agents.

When you run OpenClaw seriously — multiple agents, dozens of cron jobs, sub-agents spawning sub-agents, several Telegram groups and Whatsapp, Slack, and Discord channels, 10+ models, multiple agents and sub-agents — information gets scattered fast.

The problem: there was no single place to answer the obvious questions:

• Is my gateway actually running right now?
• How much have I spent today, and which model is burning the most?
• Which cron jobs ran, which failed, and when does the next one fire?
• What sessions are active and how much context are they consuming?
• Are my sub-agents doing useful work or spinning in circles?
• What's the cost trend over the last 7 days — am I accelerating?

The only way to answer these was digging through log files, running CLI commands, and mentally stitching together a picture from 5 different sources. That friction adds up.

The solution: a single local page that collects everything in one place — gateway health, costs, cron status, active sessions, sub-agent runs, model usage, git log — refreshed automatically, no login, no cloud, no external dependencies. Open a browser tab, get the full picture in seconds.

It's not trying to replace the OpenClaw CLI or Telegram interface. It's the at-a-glance overview layer that tells you whether everything is healthy and where your money and compute are going — so you can make decisions without hunting for data.

1. 🔔 Header Bar — Bot name, online/offline status, auto-refresh countdown, theme picker
2. ⚠️ Alerts Banner — Smart alerts for high costs, failed crons, high context usage, gateway offline
3. 💚 System Health — Gateway status, PID, uptime, memory, compaction mode, active session count
4. 💰 Cost Cards — Today's cost, all-time cost, projected monthly, cost breakdown donut chart
5. ⏰ Cron Jobs — All scheduled jobs with status, schedule, last/next run, duration, model
6. 📡 Active Sessions — Recent sessions with model, type badges (DM/group/cron/subagent), context %, tokens
7. 📊 Token Usage & Cost — Per-model breakdown with 7d/30d/all-time tabs, usage bars, totals
8. 🤖 Sub-Agent Activity — Sub-agent runs with cost, duration, status + token breakdown (7d/30d tabs)
9. 📈 Charts & Trends — Cost trend line, model cost breakdown bars, sub-agent activity — all pure SVG, 7d/30d toggle
10. 🧩 Bottom Row — Available models grid, skills list, git log
11. 💬 AI Chat — Ask questions about your dashboard in natural language, powered by your OpenClaw gateway

• 🔄 On-Demand Refresh — server.py refreshes data when you open the dashboard (no stale data)
• ⏱️ Auto-Refresh — Page auto-refreshes every 60 seconds with countdown timer
• 🎨 6 Built-in Themes — 3 dark (Midnight, Nord, Catppuccin Mocha) + 3 light (GitHub, Solarized, Catppuccin Latte), switchable from the UI
• 🖌️ Glass Morphism UI — Subtle transparency and hover effects
• 📱 Responsive — Adapts to desktop, tablet, and mobile
• 🔒 Local Only — Runs on localhost, no external dependencies
• 🐧 Cross-Platform — macOS and Linux
• ⚡ Zero Dependencies — Pure HTML/CSS/JS frontend, Python stdlib backend
• 💬 AI Chat — Natural language queries about costs, sessions, crons, and config via OpenClaw gateway
• 🎯 Accurate Model Display — 5-level resolution chain ensures every session/sub-agent shows its real model, not the default

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

This will:

1. Install to ~/.openclaw/dashboard
2. Create a default config
3. Run initial data refresh
4. Start server.py as a system service
5. Open http://127.0.0.1:8080

# Clone the repo
git clone https://github.com/mudrii/openclaw-dashboard.git ~/.openclaw/dashboard
cd ~/.openclaw/dashboard

# Create your config
cp examples/config.minimal.json config.json
nano config.json  # Set your bot name

# Start the server (refreshes data on-demand)
python3 server.py &

# Or bind to LAN for access from other devices
python3 server.py --bind 0.0.0.0 &

# Open in browser
open http://127.0.0.1:8080  # macOS
xdg-open http://127.0.0.1:8080  # Linux

Click the 🎨 button in the header to switch themes instantly — no reload or server restart needed. Choice persists via localStorage.

Add your own themes by editing themes.json. Each theme defines 19 CSS color variables:

{
  "my-theme": {
    "name": "My Theme",
    "type": "dark",
    "icon": "🎯",
    "colors": {
      "bg": "#1a1a2e",
      "surface": "rgba(255,255,255,0.03)",
      "surfaceHover": "rgba(255,255,255,0.045)",
      "border": "rgba(255,255,255,0.06)",
      "accent": "#e94560",
      "accent2": "#0f3460",
      "green": "#4ade80",
      "yellow": "#facc15",
      "red": "#f87171",
      "orange": "#fb923c",
      "purple": "#a78bfa",
      "text": "#e5e5e5",
      "textStrong": "#ffffff",
      "muted": "#737373",
      "dim": "#525252",
      "darker": "#404040",
      "tableBg": "rgba(255,255,255,0.025)",
      "tableHover": "rgba(255,255,255,0.05)",
      "scrollThumb": "rgba(255,255,255,0.1)"
    }
  }
}

server.py          ← HTTP server + /api/refresh + /api/chat endpoints
  ├── index.html   ← Single-page dashboard (fetches /api/refresh, /api/chat)
  ├── themes.json  ← Theme definitions (user-editable)
  ├── refresh.sh   ← Data collection script (called by server.py)
  └── data.json    ← Generated data (auto-refreshed)

When you open the dashboard, index.html calls /api/refresh. The server runs refresh.sh (with 30s debounce) to collect fresh data from your OpenClaw installation, then returns the JSON. No cron jobs needed.

The /api/chat endpoint accepts {"question": "...", "history": [...]} and forwards a stateless request to the OpenClaw gateway's OpenAI-compatible /v1/chat/completions endpoint, with a system prompt built from live data.json.

The entire frontend lives in a single <script> tag inside index.html — zero dependencies, no build step. The JS is organized into 7 plain objects:

┌─────────────────────────────────────────────┐
│                 App.init()                   │
│       (wires everything, starts timer)       │
└───────┬──────────────┬──────────────┬───────┘
        │              │              │
   ┌────▼────┐   ┌─────▼─────┐  ┌────▼─────┐
   │  State  │◄──│ DataLayer │  │  Theme   │
   │ (truth) │   │  (fetch)  │  │ (colors) │
   └────┬────┘   └───────────┘  └──────────┘
        │
   ┌────▼────────────┐
   │  DirtyChecker   │
   │ (what changed?) │
   └────┬────────────┘
        │
   ┌────▼────┐   ┌────────┐
   │Renderer │   │  Chat  │
   │  (DOM)  │   ���  (AI)  │
   └─────────┘   └────────┘

All inline onclick handlers route through window.OCUI — a thin namespace that calls State.setTab() / App.renderNow(). No bare globals remain outside the module objects and top-level utilities ($, esc, safeColor, relTime).

See ARCHITECTURE.md for the full specification.

Edit config.json:

{
  "bot": {
    "name": "My Bot",
    "emoji": "🤖"
  },
  "theme": {
    "preset": "nord"
  },
  "refresh": {
    "intervalSeconds": 30
  },
  "server": {
    "port": 8080,
    "host": "127.0.0.1"
  },
  "ai": {
    "enabled": true,
    "gatewayPort": 18789,
    "model": "your-model-id",
    "maxHistory": 6,
    "dotenvPath": "~/.openclaw/.env"
  }
}

The chat panel requires:

1. Your OpenClaw gateway running with the chatCompletions endpoint enabled:

   "gateway": {
     "http": { "endpoints": { "chatCompletions": { "enabled": true } } }
   }

2. OPENCLAW_GATEWAY_TOKEN set in your .env file (defaults to ~/.openclaw/.env)

The chat is stateless — each question is sent directly to the gateway with a system prompt built from live data.json. No agent memory or tools bleed in.

See docs/CONFIGURATION.md for full details.

Full dashboard view — all sections at a glance:

Real-time bot status, gateway uptime, memory usage, active session count, today's cost, all-time spend, projected monthly cost, and a per-model cost breakdown donut chart. Smart alert banners surface high costs, failed crons, and context overflows automatically.

Three always-visible SVG charts with 7d/30d toggle: cost trend over time, per-model cost breakdown bars, and sub-agent activity volume. No external chart libraries — pure inline SVG.

All scheduled jobs with status badges (active/idle/error), schedule expression, last run time, next run, duration, and the model used. At-a-glance view of your automation health.

Live sessions with model, type badges (DM / group / subagent), context usage %, and token count. Above the session list: a visual agent hierarchy tree showing parent → sub-agent → sub-sub-agent relationships with live/idle status and trigger labels — updated every refresh.

Per-model token and cost breakdown with 7d / 30d / all-time tabs. Includes input tokens, output tokens, cache reads, and total cost per model — sortable at a glance.

All sub-agent runs with cost, duration, status, and token breakdown. Separate 7d/30d tabs. Useful for tracking which tasks spawn the most agents and where spend is concentrated.

Quick reference panel showing all configured models, active skills, and the last 5 git commits from your OpenClaw workspace — so you always know what's deployed.

Full agent setup at a glance: model routing chain (primary → fallbacks), sub-agent routing by purpose (General / Dev+Coding / Work), agent details table with per-agent fallbacks, agent bindings with resolved group names, runtime config (compaction, memory flush), and subagent limits (max depth, max children/agent).

./uninstall.sh

Or manually:

# macOS
launchctl unload ~/Library/LaunchAgents/com.openclaw.dashboard.plist
rm -rf ~/.openclaw/dashboard

# Linux
systemctl --user stop openclaw-dashboard
rm -rf ~/.openclaw/dashboard

• Python 3.6+ — Backend server and data collection
• OpenClaw — Installed at ~/.openclaw (docs)
• macOS 10.15+ or Linux (Ubuntu 18.04+, Debian 10+)
• Modern web browser

See CONTRIBUTING.md for guidelines.

MIT License — see LICENSE

Made with 🦞 for the OpenClaw community