error-recovery

Timeout handling detects long-running external model calls exceeding thresholds (e.g., >30s) and triggers retries or graceful fallback strategies. Configure a timeout policy, wrap queries with timeout detection, and define fallback behavior to prevent workflows from stalling on unresponsive dependencies.

API failure recovery catches 4xx/5xx errors and network failures, then applies exponential backoff retries with configurable limits. Implement safe fallback strategies to continue execution with partial results, ensuring workflows remain resilient despite temporary API unavailability.

Exponential backoff retries implement increasing delays between retry attempts to avoid overwhelming failed services. Define maximum retry limits and backoff parameters, then apply them to error scenarios like timeouts and API failures to achieve reliable error recovery in production workflows.

Partial success strategies consolidate available results when some steps fail, allowing workflows to continue with degraded functionality. This approach handles scenarios like missing tools or out-of-credits errors while preserving completed work and enabling downstream consolidation of outputs.

User cancellation handling gracefully stops workflows mid-execution and preserves partial results for later resumption. Detect cancellation signals, interrupt external calls, persist intermediate state, and provide mechanisms to resume without losing completed work.

Production error recovery addresses external model timeouts, API failures (401/500/network), user cancellation, missing tools, out-of-credits scenarios, and partial completion cases. Categorize each error type and apply appropriate retry, fallback, or degradation strategies to maintain system resilience.