ClawMetry CLI
ClawMetry reads the session files agents already write and shows the timeline, tool calls, tokens and cost in one dashboard.
- 436 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The dashboard reads local session and log files that may hold secrets and conversation content.
- The optional cloud connection and approval hooks widen access to tool calls.
Install
- Works in
Manual install
pip install clawmetry && clawmetryRequires Python 3.8 or newer. The dashboard opens at http://localhost:8900. A desktop app exists for macOS, Windows and Linux.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The dashboard reads local session and log files that may hold secrets and conversation content.
- The optional cloud connection and approval hooks widen access to tool calls.
This is third-party code. Review the repository files before installing.
What it does
The `clawmetry` CLI starts a local dashboard at `localhost:8900` and finds supported agent runtimes on the machine by itself: 32 in total, including Claude Code, Codex, Hermes and OpenClaw. The dashboard reads the session files and logs agents already write, so no SDK and no changes to agent code are needed. The user sees a session timeline, tool-call replay, tokens and cost by runtime, model and day, signals of looping and repeated failures, context window utilization per provider and budget alerts. The default mode is observation only; controls are opt-in: the Pro plan can pause risky tool calls before they run and ask for approval from a phone. The `agent-kill-switch` skill teaches an agent to install ClawMetry and stop a session on request.
Who it is for. People running several agents who want to tell a long run that is working from one that is stuck, and see what it costs.
Good fit when
- An agent made a hundred tool calls with no visible progress.
- The token bill grew and it is unclear which session caused it.
- An agent's context hits the model window and you need to see where compaction happens.
- Several agents run on one machine and a single overview is needed.
Not a fit when
- You need Claude Code, Codex or Cursor covered for free: the free tier reads only OpenClaw, NemoClaw, Goose and Qwen Code.
- Session data cannot leave the machine even encrypted: cloud fleet connection is a separate command and the rest works locally.
- You need blocking of agent actions in every runtime: controls really work only in some.
Example request
Install ClawMetry and show what the agents on this machine are doing right now and how much they spent today.Limitations
The free version reads four runtimes: OpenClaw, NVIDIA NemoClaw, Goose and Qwen Code. The other 28, including Claude Code, Codex and Cursor, are read by the closed-source `clawmetry-pro` package on paid plans from 9 dollars per node per month. Runtimes expose different data and some publish no cost at all. Approval and blocking of calls work for only some runtimes, see the approvals document. An anonymous install ping and a PyPI version check run by default and can be turned off. Session content does not leave the machine until `clawmetry connect` is run. Python 3.8 or newer is required.
How to disable. Stop the process with Ctrl+C and remove the package with `pip uninstall clawmetry`. If you connected the cloud or installed hooks and MCP, remove them following the project documentation.
FAQ
How do I install ClawMetry?
Run `pip install clawmetry && clawmetry` and open `http://localhost:8900`. A desktop app exists for macOS, Windows and Linux.
What does the demo show without agents?
The `clawmetry --sample` command opens the dashboard on three labelled synthetic sessions.
Which runtimes are free?
OpenClaw, NVIDIA NemoClaw, Goose and Qwen Code. The other 28 are read by the paid `clawmetry-pro` package.
What leaves the machine?
Session content stays until you run `clawmetry connect`. An anonymous install ping and a version check run by default; they are listed in the EGRESS document.
How much overhead does it add?
Reading session files adds nothing to the agent. The HTTP interceptor, if enabled, adds about 0.44 ms per model call.
Discussion
No questions or reviews yet.
Related
CLI toolsA CLI that connects local agents to a shared cloud store of sessions, memory, skills and MCP connectors.
The clawctl CLI manages a fleet of AI agents on local machines over SSH and Ansible: install, start, upgrade and token tracking.
One runtime for Claude Code, Codex, OpenCode and more: a run counts as done only when checks the agent did not write pass
The clawdash terminal panel shows an OpenClaw gateway: sessions, schedule, activity feed and log on one screen, read-only
A local real-time dashboard for Claude Code and Codex sessions: agent activity and tool calls
MCP server of the self-hosted EmperorClaw panel: Claude Code and Codex read company knowledge, tasks and customers in scope
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| ClawMetry (this card) | no | 436 | |
| claude-dashboard | no | 587 | a dashboard for Claude Code; ClawMetry covers dozens of runtimes and prices them per provider |
| AgentOps | no | 454 | tracing through an SDK in code; ClawMetry reads existing session files with no changes |
README in short
The author lists the limits openly: runtimes expose different data and observing an action is not the same as being able to block it. Overhead was measured on an Apple M2 Pro, and the daemon's overshoot of its own budget is published as a bug.