AgentDog Plugin for OpenClaw

Monitor your OpenClaw agents with AgentDog — real-time observability, per-tool permission controls, and human-in-the-loop approval flows.

Why AgentDog?

AI agents are powerful — but you need to stay in control. AgentDog gives you:

• 🔍 See everything — every message, tool call, and cost in real-time
• 🛡️ Set boundaries — allow, block, or require approval for any tool
• ✅ Stay in the loop — approve dangerous actions with one tap, right in chat
• ⏳ Know what's happening — live activity feed shows every step as it runs
• 🛑 Stop anytime — one tap halts all agent actions instantly

Setup

1. Install

npm install -g @agentdog/openclaw

2. Get an API key at agentdog.io/settings

3. Add to openclaw.yaml

extensions:
  agentdog:
    apiKey: YOUR_API_KEY
    permissionsEnabled: true

4. Restart your agent — done!

Permission Controls

🛡️ Tool-Level Rules

Set rules for any tool in the dashboard:

• Allow — tool runs freely
• Block — tool is always prevented
• Ask me — agent pauses and sends you approve/deny buttons

Rules support exact names (send_email), wildcards (github.*), or catch-all (*). Scope them per-agent or globally.

Bulk rule creation: Select multiple capabilities and set permissions in one click.

✅ Inline Approval Buttons

When a tool needs approval:

⚠️ Approval needed

Your agent wants to run: exec
  {"command": "git push origin main"}

[✅ Approve]  [❌ Deny]

• Non-blocking — the agent continues chatting while you decide
• One-tap — tap a button, message updates to show your choice
• Text fallback — reply approve or deny if buttons aren't available
• 10-minute expiry — auto-denies if no response

🔍 Smart CLI Mapping

AgentDog automatically maps CLI commands to apps for smarter rules:

Auto-discovery: Unknown CLI tools are automatically detected and their capabilities populated from public documentation.

In-Chat Observability

⏳ Live Activity Feed

Every tool call your agent makes shows up in a single, auto-updating message — right in your chat:

⏳ exec — git pull origin main...
[🛑 Stop]

As the agent works through multiple steps, the message builds up in real-time:

✅ exec — git pull origin main (2.1s)
✅ exec — npm run build (12.4s)
⏳ exec — docker-compose up...
[🛑 Stop]

When the turn completes, you get a summary:

✅ exec — git pull origin main (2.1s)
✅ exec — npm run build (12.4s)
✅ exec — docker-compose up (8.3s)

📊 3 tools · 22.8s total

• One message, always updating — no spam, no clutter
• Every step visible — know exactly what your agent is doing
• Timing for each step — spot slow operations instantly
• Error tracking — failed steps show ❌ with error counts in the summary

🛑 Emergency Stop

The stop button is always there while the agent is working. One tap and:

🛑 Agent stopped
All tool calls are blocked until you resume.

[▶️ Resume]

• Blocks all subsequent tool calls — the agent can't do anything until you say so
• Resume when ready — tap ▶️ or just send a new message
• Auto-expires after 5 minutes (safety net)
• No data loss — conversations and context are preserved

Dashboard Observability

The AgentDog dashboard gives you the full picture:

No secrets or API keys are ever sent.

Supported Channels

All features work on:

• Telegram ✅
• Discord ✅
• Slack ✅

Other channels fall back to text-based interactions or the dashboard.

Config Options

extensions:
  agentdog:
    apiKey: ad_xxxxxxxxxxxxx        # required
    endpoint: https://agentdog.io/api/v1  # optional
    agentName: my-agent             # optional, shown in dashboard
    syncInterval: 86400             # config sync interval in seconds
    permissionsEnabled: true        # enable permission enforcement

Links

• Dashboard: agentdog.io
• Permissions: agentdog.io/permissions
• OpenClaw: openclaw.ai
• Issues: GitHub

Built with 🐕 by AgentDog