Cost & usage
Every Firetrail analysis is a set of AI calls, and AI calls cost real money. Most products hide that; we'd rather show you. The Cost & Usage page tracks what a project's analysis actually costs, estimated from the tokens each AI call consumed, priced at the models' public rates.
Why show this at all? Two reasons. If you're cost-conscious, you can see exactly where the money goes and trim deliberately. And when you're deciding whether Firetrail is worth it, "what does it cost" deserves as straight an answer as "what does it save".
The numbers
- This month — spend so far, and how many analyses produced it.
- Projected this month — where the month lands at the current pace.
- Average cost per analysis — the per-PR unit cost, with a breakdown of the calls inside it. Typically a small number of cents; a project with unusually large PRs will see it higher.
- All-time — total spend and total tokens since the project started.
The charts
- Cost over time — weekly spend. Spikes usually just track busy weeks; a rising per-analysis cost is the more interesting signal (often: PRs getting bigger).
- Cost by classifier — which checks cost the most. Costs differ because checks run on different AI models — a few of the hardest-thinking checks use a stronger model, and the rest run on a fast, inexpensive one.
Things that affect your cost
- Each active custom classifier adds one AI call to every analysis. A handful is negligible; dozens on a high-volume repo will show up here.
- Cached re-runs are free. Re-analyzing a PR whose commit hasn't changed reuses stored results and makes no AI calls at all.
- PR size matters. Bigger diffs mean more tokens per call — one more quiet argument for smaller PRs.