OpenClaw agents make LLM API calls continuously — user messages, heartbeat
timers, cron jobs, subagents. Without visibility, it's easy to spend $50–200/mo
without knowing where the money goes.
CostClaw captures every call via OpenClaw's native event hooks and serves
a local dashboard at localhost:3333. No external service, no account, all
data in a local SQLite file.
What it shows:
- Per-model cost breakdown with share %
- Cost by trigger type (heartbeat spend is almost always the surprise)
- Per-session costs (group by conversation)
- Hourly view for today, daily for 7d/30d
- Auto-generated recommendations based on actual usage
GitHub: https://github.com/Aperturesurvivor/costclaw-telemetry
Built in TypeScript with better-sqlite3 and Chart.js. MIT licensed.
The most common finding: heartbeat agents running on Sonnet/GPT-4o for keep-alive checks that could run on Haiku/mini for 1/20th the cost.
OpenClaw agents make LLM API calls continuously — user messages, heartbeat timers, cron jobs, subagents. Without visibility, it's easy to spend $50–200/mo without knowing where the money goes.