Cost levers

This page is about what a run costs to serve. Customers pay a fixed plan price; see Runs and allowances for what they see.

The crawl is nearly free. The model calls are the cost, and they scale with pages and intents. Measured on a full-profile run: about $17 for 500 pages, $80 for 2,500, $310 for 10,000. Batched, half that. npm run estimate -- 500 2500 10000 prints the table for the current profile.

Levers, in order of leverage

1. Analysis profile — test runs every step on the fast model at low effort and caps graph extraction at 8 pages and judged intents at 20. Use it while iterating on a site. full for the monthly report. Set per workspace in Settings → General.
2. Batch — with LLM_BATCH=true, the two bulk steps (page-graph extraction, intent judging) go through the Batches API at 50% cost with minutes-to-hours latency. Right for scheduled runs, wrong for watching a run live.
3. Incremental runs — a page whose content is unchanged since the last run keeps its graph and claims instead of being re-extracted. On a stable site the monthly run touches only what changed.
4. Page budget — extraction cost is linear in pages. Template findings stabilise at a few hundred; spend more only for long-tail coverage.
5. Token budget — a per-run ceiling. A run that exceeds it pauses rather than failing; raise it and resume.

Where to look

Usage and cost shows month-to-date spend against the plan budget and cost per run. Every model call, page fetch and render is a metered event with an estimated price.