Rate limits
Per-key and per-organization limits, and how to handle 429.
Batchwell applies sliding-window rate limits per key and per organization. They are checked before any credit is held or any model runs, so a rate-limited request costs nothing.
When you hit a limit you get HTTP 429 with a Retry-After header in seconds. Wait at least that long before
retrying, and add jitter if several workers share a key.
Some models are served by providers with small concurrency limits. When an endpoint is busy we skip it for that
attempt and try the next eligible one. If none is free you may see 503 no_eligible_endpoint; add
fallback models to reduce this.
Write to support@batchwell.ai with your organization name and expected request rate.