Self-Healing Automations

What self-healing means

Chase Agents can automatically attempt to fix failed automations using AI. When an automation fails, the platform analyzes the error, generates a corrected version of the failing step, and retries. This reduces manual intervention for recoverable errors.

Error handling modes

• any — attempt AI auto-fix for any error type
• ai — attempt AI auto-fix only for errors caused by AI reasoning mistakes (wrong field names, bad data transformations)

Pre and post execution auto-fix

• Pre-execution auto-fix — reviews steps before running and fixes obvious issues before they cause errors
• Post-execution auto-fix — when a run fails, generates a fix and automatically retries

What auto-fix handles well

• Incorrect field names in data references
• Python syntax errors and simple logic mistakes
• Incorrect parameter names in API calls

What requires manual intervention

• Expired or revoked API credentials
• External service outages
• Breaking changes to an API's response schema
• Rate limit violations requiring structural changes such as adding wait steps

Success criteria

Each automation can have a success_criteria string in plain language describing what a successful run looks like. For example: 'Slack message sent with customer count greater than zero.' This helps the auto-fix system and run evaluation logic determine whether a run truly succeeded beyond just completing without errors.