Rate Limits

v1 limits are deliberately permissive on reads and explicit on paid surfaces.

Read-only MCP calls

No per-call rate limit in v1. Cloudflare's edge applies basic abuse limits at the platform level. If you find yourself making thousands of identical calls, cache the resource (aicoach://models/catalog, skill markdown) on your side.

Free-tier nano apps

3 runs/day per tool per user. Counter resets at UTC midnight. Once exceeded, the tool returns payment_required and the user can either wait or top up.

Model bake-offs (when shipped)

• Free: 5/day, ≤3 models per call, ≤8k output tokens.
• Paid: any model in the catalog, up to 6 per call, billed at catalog rate × markup.

Per-key spend caps

Each API key supports daily and monthly cap fields (USD cents). Exceeded caps hard-stop paid calls with payment_required until the period resets.

Set them at issue-time on /account/api-keys. We strongly recommend setting a cap on any key used by an autonomous agent.