How to Handle AI Errors and Rate Limits in Make.com Scenarios
Add retries, breaks, and fallbacks so your AI scenarios survive rate limits and bad responses without breaking.
AI modules fail in predictable ways: rate limits, timeouts, and malformed output. A scenario that ignores this stops at the worst moment. This guide adds the error handling that keeps your AI flows running unattended.
- An existing Make scenario that calls OpenAI
- A Make.com account with scenario editing access
- A place to log failures, such as a Slack channel or a sheet
Step 1: Understand the common errors
The two you will see most are HTTP 429 (too many requests, a rate limit) and a JSON parse failure when the model returns prose instead of clean JSON. Each needs a different response: retry the first, sanitize or skip the second.
| Symptom | Likely cause | Fix |
|---|---|---|
| 429 error | Rate limit hit | Break directive with retry |
| Timeout | Slow model response | Retry with longer interval |
| Parse JSON failed | Model returned extra text | Strip fences, then Resume |
| Empty result | Bad or blocked prompt | Ignore and log |
Step 2: Add a Break handler for rate limits
Right click the OpenAI module and choose Add error handler, then add a Break directive. Break tells Make to wait and retry the failed bundle automatically. Set the number of attempts and the interval between them.
Step 3: Configure retry settings
Open the scenario settings and enable Sequential processing so retries do not pile up. On the Break handler set retries to about 3 and the interval to a few minutes, which clears most short rate limit windows.
Error handler: Break
Number of attempts: 3
Interval between attempts: 5 minutes
Scenario settings:
Sequential processing: On
Allow storing of incomplete executions: OnStep 4: Add a Resume fallback for bad output
For parse failures, add a Resume directive on the Parse JSON module. Resume lets the scenario continue using a default value you supply, so one malformed response does not kill the whole batch. Route those cases to a review log.
Step 5: Log every failure
On the Ignore or Resume path, add a Slack or Sheets module that records the input that failed and the error message. You want visibility into what slipped through so you can refine the prompt later.
Result: a resilient AI scenario that quietly retries rate limits, recovers from bad responses with a fallback, and tells you about anything it could not handle, instead of silently stopping.
Watch related tutorials
25:00
12:05
18:30
32:08
21:45
28:30