Systematic debugging approach
When diagnosing an issue, follow this sequence:
- Check Observability: Open Observability for the exact payload, provider sequence, and error details
- Inspect the timeline: The timeline view shows the complete request lifecycle, including failed provider attempts, webhook deliveries, and callbacks
- Verify provider quotas: Confirm provider-side quotas or rate limits haven't been reached
- Isolate the issue: Re-run the request with a known-good workflow to determine whether the issue is payload-related or system-wide
- Check customer data: Use cached data to trace which end user triggered the request
Using Observability for debugging
Request Logs
- Filter by workflow, status, provider, or date range
- Click into any request to see the full payload and response
- Check
meta.attemptsto see if fallback providers were used
Timeline view
The timeline provides a chronological view of:
- Provider selection and request dispatch
- Fallback attempts with failure reasons
- Webhook delivery attempts and responses
- Event-driven callbacks with processing times
Payload inspection
- Compare the sent payload against your expected schema
- Verify
messagesarray format matches provider expectations - Check that cache fields reference valid paths in your payload
Common debugging scenarios
AI returns unexpected format
- Check if a structured output is attached to the workflow
- Verify the schema matches your expectations
- Try a more capable model (e.g., upgrade from gpt-3.5 to gpt-4o)
- Review the raw provider response in Request Logs
Fallback provider keeps activating
- Check the timeline for the primary provider's failure reason
- Verify primary provider credentials are valid
- Monitor primary provider's status page for outages
- Consider adjusting timeout thresholds
Webhook not delivered
- Check the webhook delivery status in Request Logs
- Verify the endpoint URL is accessible and returns
2xx - Check for signature verification failures in your server logs
- Review Delivery & retries for retry status
For comprehensive guidance on using request logs for debugging, see the Observability documentation.
Next steps
- Contact support: Get help from the ModelRiver team
- Quick fixes: Common error solutions
- Observability: Full monitoring documentation