How it works

Full visibility into
every AI agent action

ClawMetry sits alongside OpenClaw and continuously monitors every tool call, sub-agent, memory read, and token spent. When something goes wrong, you know immediately.

Install ClawMetry See it in action

Where ClawMetry fits

No proxy. No code changes. ClawMetry reads OpenClaw's output stream and maps it into a live dashboard with zero overhead on your agents.

🤖
Your AI Agents
Running on OpenClaw
Any model, any tools
tool calls
+ events
🦞
ClawMetry
Real-time monitoring
Alerts + dashboard
alerts
+ control
📊
Live Dashboard
🔔
Slack / PagerDuty
Human Approval

Up and running in 3 steps

From zero to full observability in under a minute.

1

Install

One pip install, one command to connect. ClawMetry auto-discovers your OpenClaw gateway and starts listening immediately.

# Install
pip install clawmetry
# Connect to OpenClaw
clawmetry connect
✓ Connected to OpenClaw gateway
✓ Dashboard at localhost:8900
2

Monitor

ClawMetry tails every event your agents produce. No instrumentation required. Works with your existing setup, unmodified.

Every tool call (name, input, output, duration)
Sub-agent lifecycle (spawned, running, done/failed)
Memory reads and writes
Token usage and cost per session
Cron jobs: scheduled, fired, missed
Error patterns and unusual loops
3

Act

When something looks wrong, you have full control. Don't just observe - intervene.

🔔
Get alerted via Slack, PagerDuty, Telegram, or email the moment something fails
⏸️
Pause a running agent mid-task to inspect its state
Approve or deny sensitive actions before they execute
🔁
Resume or redirect an agent once you've reviewed

Everything your agents do, visible

ClawMetry tracks the full lifecycle of every agent session - from first tool call to final output.

🔧

Tool calls

Every tool invocation: name, inputs, outputs, latency, and whether it succeeded or failed. Search and filter by tool type or session.

🤖

Sub-agent lifecycle

When sub-agents are spawned, what tasks they run, how long they take, and how they exit. No more blind spots in multi-agent chains.

💰

Token usage and cost

Per-session, per-model, per-tool cost breakdown in real time. Know exactly which agents are burning your budget before the invoice lands.

⚠️

Errors and failures

Tool errors, retries, stuck loops, and silent failures surface immediately in the dashboard - and trigger alerts to your team.

🧠

Memory activity

Every memory file read and write, with before/after diffs. See exactly how your agents' long-term context changes over time.

Cron jobs

Scheduled agent tasks with run history, success rates, and last-fired timestamps. Catch missed runs before they become incidents.

When things go wrong,
you'll know immediately

Not 3 hours later when you check your email. ClawMetry fires alerts the moment something goes wrong - to wherever your team already lives.

🔔
PagerDuty & OpsGenie
Page your on-call. Works with your existing runbooks.
💬
Slack & Discord
Post to a channel, tag the right person.
✈️
Telegram
Native via OpenClaw. Zero extra config.
📧
Email
For when async is fine and a page isn't needed.
Agent silent for >10 minutes
No tool calls or responses. Could be stuck waiting, blocked on I/O, or silently crashed.
Sub-agent never completed
Spawned but no exit event after a configurable timeout. Common in broken tool chains.
Tool call failure rate above threshold
More than N% of tool calls erroring in the last rolling window. Often signals a broken API or bad input.
Cost spike detected
Token usage exceeds your defined budget per session or per time window. Catch runaway loops before they drain your API credits.
Unusual loop pattern
Same tool called more than N times in a row, or an identical input/output pair repeating. Agent is probably stuck.
Agent: file-processor
⚠️
Sensitive action detected
Agent is about to call exec() with command: rm -rf /tmp/cache/*
Waiting for approval
Agent is paused. Review the action and approve or deny before it proceeds.
Approve
Deny
Approved - agent resumed
Command executed. Agent continuing with next step.

You stay in the loop

Observability without control is just logging. ClawMetry lets you pause, inspect, and approve what your agents do - before it's too late to reverse.

⏸️

Pause any running agent

Interrupt a live session mid-task. The agent holds its state until you're ready to continue.

🔍

Inspect before acting

See the exact tool call, inputs, and context the agent was about to use. Make the call with full information.

Approve sensitive actions

Flag specific tools or patterns that require human sign-off. Agents wait. You decide.

Ready to see what's happening?

Free, open source, and running in under a minute. No account required.

pip install clawmetry && clawmetry connect