One dashboard for 30 agent runtimes
ClawMetry reads the files each agent runtime already writes, so there is no SDK, no proxy and no code change. Every runtime below has its own page describing exactly what you get: which sessions, tool calls and tokens appear, whether cost is measurable, and whether a session can be paused or stopped.
Each card links to that runtime's page. The one-line summary is the honest scope, not a slogan: where a runtime keeps no cost on the machine, the page says so and ClawMetry reports spend as unavailable rather than guessing.
Every runtime, one line each
- OpenClawSessions, channel messages, cron runs and tokens. Dashboard and mission control. Free tier.
- Claude CodeUsage monitor and cost dashboard: every session, tool call, sub-agent tree and dollar.
- CursorUsage dashboard for Cursor agents: chats, tool calls and token cost, no IDE extension.
- GitHub CopilotUsage dashboard: every Copilot chat, tool call and token, read locally.
- CodexCodex CLI usage monitor: sessions, tool calls and cost per session.
- Gemini CLIUsage dashboard: sessions, tool calls, sub-agents and cost per session.
- LovablePer-edit activity timeline from the GitHub-synced clone. No cost figures, no live indicator, observe only.
- ClineEvery task, tool call, subagent and token, with cost per task across providers.
- OpenHandsSessions, tool calls, delegated sub-agents and cost per session.
- OpenWorkerSessions, connector actions, approvals and cost per session.
- OpenCodeSessions, tool calls and tokens as they happen.
- NVIDIA NemoClawObservability inside OpenShell sandboxes: every claw, token cost and memory file. Free tier.
- AiderEvery chat, file edit and token.
- GooseSessions, tool calls and tokens for Block Goose. Free tier.
- n8nAI workflow runs, node calls and tokens; cost where the model node records usage.
- Qwen CodeSessions, tool calls and tokens for the Qwen Code CLI.
- Devin CLISessions, tool calls, subagents and tokens, local-first. Devin Cloud and Devin Desktop are not observed.
- AntigravityUsage monitor: every session, tool call and Gemini token.
- Grok BuildSessions, tool calls and tokens for the xAI CLI coding agent.
- Grok BotConversations and the local permissions each bot holds. Spend is not observable from the client.
- Hermes AgentDashboard, observability and governance: sessions, tool calls and tokens.
- Pi coding agentSessions, tool calls and tokens.
- Deep AgentsLangChain Deep Agents runs, tool calls and tokens.
- Kimi CLITurns, tool calls, subagents and tokens, with cost per session.
- DeepSeek HarnessSessions, tool calls and tokens for dsh.
- ExoConversations, forks, tool calls and native per-call cost.
- NanoClawDashboard with observability and governance: sessions, tool calls and tokens.
- PicoClawSessions and per-provider token cost.
- QMTurns, scopes and tokens, with no change to your qm code.
- Replit AgentFull session transcripts, tool calls, tool errors and the progress checklist. Cost is not stored in the workspace, so none is claimed.
Three scope notes worth reading before you pick
- Cost is only claimed where it is measurable. Lovable bills credits in its own cloud, Replit Agent bills effort-based checkpoints server-side, and Grok Bot runs its model on an xAI-hosted VM. None of them leave a meter on your machine, so those pages report spend as unavailable instead of zero.
- Control varies by runtime. A Cursor CLI session is a real process and can be paused or killed; a Cursor editor conversation shares the IDE process and cannot. OpenClaw has no pause primitive, so a pause is advisory unless the enforcement proxy is in front of the model calls. Every page states its own answer.
- Free and paid tiers. OpenClaw, NVIDIA NemoClaw and Goose are free on every plan; the rest are on the paid tiers listed on the pricing page.
Missing yours? The supported-runtimes strip on the homepage has an "Add yours" request that goes straight to the roadmap.
One command, every runtime on this machine.
pip install clawmetry && clawmetry, then open localhost:8900. Nothing to configure: the runtimes are detected from the files they already write.
See ClawMetry →
← Back to ClawMetry