Production Readiness

Before Production

Use bounded requests, explicit timeouts, and recoverable job handling.

1. Keep API keys in a secret manager.
2. Set result, page, depth, path, and timeout limits.
3. Request only required output formats.
4. Poll asynchronous jobs with a capped interval or use webhooks.
5. Honor Retry-After on 429.
6. Retry network and 5xx failures with backoff and jitter.
7. Validate critical extracted fields before consequential use.
8. Monitor credits, concurrency, latency, and failure rates.

Retry Decisions

Async Job Loop

start job → store job_id → poll with backoff → stop on completed, failed, or cancelled