Rate limits & usage
Requests, reasoning units (RU), caching, rate limits, and retries.
Rate limits
Limits are per API key, per minute. 429 responses include retry hints. Official SDKs retry automatically with backoff.
Reasoning units (RU)
A Reasoning Unit (RU) measures how much work a request performs. RUs power the Usage dashboard — they are not billing line items by themselves. Every authenticated /v1/* request records RUs alongside latency and status.
Non-reasoning endpoints cost 1 RU. POST /v1/reason starts at 5 RU plus a goal modifier:
| Goal | Modifier | Total RU |
|---|---|---|
| FEASIBILITY | +1 | 6 |
| TRANSFORM | +2 | 7 |
| RECOMMENDATION | +2 | 7 |
| ANALYSIS | +2 | 7 |
Caching & retries
Repeated reasoning for the same recipe + pantry may hit server-side caches. Official SDKs retry 429, 502, 503, and 504. Prefer idempotent reads when retrying POSTs in custom clients.