AI Auto-Fix: How Self-Healing Automations Work

One of Chase Agents' most powerful reliability features is AI-driven auto-fix: the ability for the platform to detect a failing step, understand why it failed, rewrite it, and continue execution without requiring you to manually intervene. This article explains how the auto-fix engine works and when to enable it.

What Auto-Fix Does

When a step fails and the automation's error handling mode is set to ai, the platform captures the full error context: the step definition, the inputs it received, the error message, and the stack trace if available. It passes this context to an AI model specialized in step repair. The model produces a corrected version of the step and the platform re-executes it.

If the corrected step succeeds, the automation continues from where it left off. If the correction attempt also fails, the run is marked as failed and logged for manual review. The auto-fix engine does not attempt infinite retries; it makes one repair attempt per failure.

Pre-Execution Auto-Fix

When automation_pre_execution_auto_fix is enabled, the platform reviews your automation's configuration before each run begins. It checks for issues like references to step outputs that do not exist, missing required parameters, and malformed expressions. If it finds problems, it attempts to correct them before execution starts. This prevents entire runs from being wasted on errors that could have been detected upfront.

Post-Execution Auto-Fix

When automation_post_execution_auto_fix is enabled, the platform analyzes a failed run after it completes. It identifies the root cause of the failure, generates a corrected version of the automation, saves the updated definition, and triggers a new run with the repaired configuration. The original failed run remains in history so you can compare the before and after.

What Auto-Fix Is Good At

The auto-fix engine excels at a specific class of problems: API response format changes where an endpoint starts returning a different JSON structure; credential expiry where a token needs to be refreshed and the step code needs to handle the new authentication flow; minor logic errors in run_sandboxed_code steps where the Python is almost correct but has a small syntax error or off-by-one bug; and parameter name changes where an external service has updated its API and renamed a field.

What Auto-Fix Cannot Fix

Auto-fix is not a solution for fundamental automation design flaws. If an automation is trying to accomplish something that is not possible with the connections it has, or if the core logic of the workflow is incorrect, auto-fix will fail repeatedly. In these cases, the run will be flagged for manual review and you will need to revisit the automation design.

Enabling Auto-Fix

Both pre-execution and post-execution auto-fix settings are configurable per automation. You can enable or disable them independently from the automation's settings panel. The error handling mode (ai versus any) is also set per automation. Workspace admins can view and change these settings for any automation in the workspace.

Reviewing Auto-Fix Decisions

Whenever auto-fix modifies a step, the change is recorded in Workflow History. You can see what the step looked like before and after the fix, what error triggered the repair, and whether the repaired step succeeded. Reviewing these records periodically helps you understand where your automations are fragile and guides manual improvements to the automation design.

Auto-Fix and Security

The auto-fix engine never modifies your credential secrets or connection configurations. It only rewrites the step logic: the code, the parameters, and the data transformation expressions. Your API keys, OAuth tokens, and other credentials remain in the secure credential vault and are never exposed to or modified by the auto-fix process.