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.