Overview
Model steps support an optional fallback that automatically retries with an alternative model if the primary model fails. This keeps your workflow humming along even during temporary outages or rate limits.
How It Works
- Execute the primary model with configured parameters
- If primary fails, automatically execute the fallback model
- If both fail, mark the step as failed
Configuration
Add a fallback object to any model step:
Fallback Properties
Detecting Fallback Usage
When a fallback is used, the step metadata includes:
Best Practices
- Choose fallback models that produce compatible output formats
- Use fallback for critical production workflows
- Monitor
fallback_used in execution results to identify recurring primary failures
- Consider cost differences between primary and fallback models
Last modified on March 6, 2026