What it does
ClawMetry is a monitoring tool for AI agent runtimes. It reads the session files and logs that agents already leave on your machine and puts them in one dashboard, so you can tell a long run that is making progress from one that is stuck. It needs no SDK or code changes, and the README says it reads those files read-only. After pip install clawmetry and running clawmetry, the dashboard opens at localhost:8900. It covers sessions and transcripts with replay, token and cost breakdowns by runtime, model, session and day, a live flow diagram, context-window utilisation, health and logs, and alerts sent to Slack, Discord, PagerDuty, Telegram or email. Pausing risky tool calls for approval is opt-in and ships switched off. It supports 33 agent runtimes. The open-source app reads OpenClaw, NVIDIA NemoClaw, Goose and Qwen Code without an account, while the others, including Claude Code, Codex and Cursor, are read by a closed-source companion that comes with a 7-day trial or a paid plan. No session data leaves the machine unless you run clawmetry connect. An anonymous install ping and a PyPI version check run by default and can be turned off. It needs Python 3.8+ on macOS, Linux or Windows.
Capabilities
- Session timelines and tool-by-tool replay across 33 agent runtimes
- Token and cost breakdowns by runtime, model, session and day
- Context-window utilisation sized per provider
- Alerts for budget caps, error spikes and offline agents
- Opt-in approval gate for risky tool calls
- OpenTelemetry export on the Pro plan
When to use it
- You run several coding agents and want one view of what they did and what they cost
- You need to see why a long agent run looks stuck
- You want cost monitoring without adding an SDK to your agent code
