For the complete documentation index, see llms.txt. This page is also available as Markdown.

Recipe: retry and backoff

Recipe — retry and backoff. Pointer to the SDK's built-in retry behaviour.

The SDK retries automatically on these conditions, no client code needed:

  • RateLimitError (429) — respects Retry-After response header

  • InternalServerError (500), 502, 503, 504 — exponential backoff

  • APIConnectionError — transient network blip

Defaults: max_retries=2, exponential backoff, configurable via Stratix(max_retries=N, timeout=...). To disable: Stratix(max_retries=0).

For per-request override: client.with_options(timeout=300.0).evaluations.create(...).

When to add your own retry

If you need request-level retries beyond the SDK's defaults (e.g., long-running batch with custom backoff), wrap the call in tenacity or similar — but most paying-customer use-cases don't require this.

See also

Last updated

Was this helpful?