Daily OpenClaw model and token usage reports from local session logs, with both a CLI and a tidy local dashboard.
This is for normal OpenClaw users who want a simple answer to questions like:
- what used GPT today?
- did Ollama Cloud get used?
- were local models active?
- did my setup drift away from local-first?
It is meant to be lightweight and practical. It is not a fancy analytics platform.
Run:
bash scripts/run_dashboard.shThen open the local page shown by the server.
Dashboard mode gives you:
- headline totals
- drift warnings
- top models
- provider breakdown
- by-model table
- refresh from the page
Single day:
python3 scripts/usage_rollup.py --date 2026-04-16 --openclaw-home ~/.openclaw --csvToday helper:
bash scripts/usage_today.shDate range:
python3 scripts/usage_rollup.py --start-date 2026-04-14 --end-date 2026-04-16 --openclaw-home ~/.openclaw --csvClaw Usage Rollup scans local OpenClaw session and cron artifacts and builds usage summaries by:
- model
- provider
- agent
- date or date range
It writes:
- JSON reports
- Markdown summaries
- CSV exports
- policy and drift summaries that help spot GPT-heavy or cloud-heavy days
This is a local artifact rollup, not billing truth.
Important limits:
- event counts are not guaranteed to equal unique model calls
- token totals depend on what providers write into local artifacts
- some providers or local models may show activity with zero usable token totals
- cost is intentionally not estimated in v1
By default it scans:
~/.openclaw/agents/*/sessions/*.jsonl*~/.openclaw/cron/runs/*.jsonl*
It ignores lock and temp files.
By default the script redacts absolute paths in its report output.
Use --show-paths only for local debugging.
Examples:
reports/model-usage/daily-2026-04-16.jsonreports/model-usage/daily-2026-04-16.mdreports/model-usage/daily-2026-04-16-by-model.csvreports/model-usage/daily-2026-04-16-by-agent.csvreports/model-usage/range-2026-04-14_to_2026-04-16.md
Best fields to trust first:
nonzero_token_requestsrecorded_total_tokens- top token usage summary
- policy summary and policy warnings
Treat these more carefully:
activity_eventstoken_bearing_requests- any model with lots of
zero_token_records - policy warnings, which are guidance rather than hard billing truth
docs/README-dashboard.mddocs/README-cli.mdSUPPORT.md
See fixtures/sample-report-daily.json for a small example report shape.
Claw Usage Rollup is free to use and will remain free to use.
There is no paywall, no locked feature tier, and no pressure to pay. If it saves someone time and they want to help with future improvements, they can optionally support it here:
Support is never required.
MIT