Cost & usage
Route: /cost
The Cost page shows LLM token consumption and estimated spend across all AI pipelines for the selected project and time range.
What’s tracked
| Pipeline | Models logged |
|---|---|
| fix-worker | Primary model (Anthropic / OpenAI), per fix_attempts row |
| classify-report | Classifier model, per classifications row |
| judge-batch | Judge model, per judge_results row |
| inventory-propose | Inventory model, per inventory_proposals row |
| pdca-runner | Planner model, per pdca_runs row |
| test-gen-from-report | Test generator model |
Each row records model, prompt_tokens, completion_tokens, and cost_usd (estimated
using public pricing at time of logging).
Charts
- Daily spend — bar chart of estimated USD per day
- By pipeline — donut chart showing which workflows consume the most tokens
- By model — which models are used and what they cost
- Token efficiency — completion tokens per successful fix (lower is better)
Controlling costs
| Lever | Where to set it |
|---|---|
| Model selection | Settings → PDCA settings → Model (use a smaller model for planning) |
| Max iterations | Settings → PDCA settings → Max iterations per run |
| Fix worker retries | fix-worker retries on Zod validation failure — reduce max_retries in env |
| Integration health probe cadence | Probes run on demand; avoid automating frequent polling |
Export
Click Export CSV to download the raw cost log for the selected date range. Useful for chargeback / showback reporting in multi-team setups.
Related pages
- Iterate (PDCA) — the highest per-run cost driver
- Fix orchestrator — fix-worker spend
- Settings — model and iteration configuration
Last updated on