429 with a Retry-After header (seconds).
Recommended pacing
| Path | Suggested RPM | Notes |
|---|---|---|
POST /v1/chat/completions | ≤ 60 | Streaming calls count too; wait for one to finish before the next |
POST /v1/files | ≤ 30 | Batch where possible; OCR is compute-heavy |
POST /v1/data | ≤ 120 | Batch records into one request (up to 1000/batch) |
GET reads (/v1/data, /v1/files, /v1/models) | ≤ 120 | — |
Exact thresholds are injected per deployment. Email Mirobody Support to confirm the limits for your account before going live.
Client backoff
- On
429, respectRetry-After— never retry immediately. - Double the backoff after two consecutive
429s, up to 5 minutes. - Don’t fire chat calls in parallel; let each stream complete first.