Context Length Exceeded: Reduce the Request Without Losing the Answer
Measure every context component, remove duplicates and irrelevant history, compact tool results, reserve output space, and verify answer quality.
Errors & Reliability
API Timeout: Find the Failing Layer and Retry Safely
Diagnose connect, read, deadline, proxy, gateway, and upstream timeouts, then verify recovery with controlled short and long requests.
API 429 Too Many Requests: Limits, Retry-After, and Safe Retries
Diagnose API 429 responses, distinguish temporary limits from quota conditions, implement bounded backoff with jitter, and avoid duplicate side effects.
Streaming and SSE Disconnects: Recover Without Duplicates
How to distinguish a terminal event from an SSE disconnect, preserve partial output, inspect timeouts, and retry without duplicate side effects.
Model Not Found in an API: How to Locate and Fix the Cause
A diagnostic order for endpoint, protocol, API key, current Model ID, aliases, client overrides, status, and request ID.