openclaw - 💡(How to fix) Fix Bug: Silent failure and fallback override when OpenAI usage limit (429) is hit [3 comments, 1 participants]

When the OpenAI usage limit or rate limit (HTTP 429) is reached, OpenClaw detects the error but fails to proactively notify the user via configured channels (e.g., Slack). Furthermore, although the system correctly selects the fallback model (e.g., google/gemini-3.1-pro-preview), a live session model state overwrite occurs right before the execution attempt, forcing the model back to the rate-limited openai-codex/gpt-5.4. 4. Observe that no error notification is sent to the user's client (e.g., Slack).

• A clear error message or notification should be routed to the user's active channel (e.g., Slack) indicating the model limit was reached.

TL;DR

Implement a mechanism to prevent live session model state overwrites when a fallback model is selected due to rate limiting or usage limits.

Guidance

• Review the logic for handling rate limit errors (HTTP 429) to ensure that notifications are properly sent to configured channels like Slack.
• Investigate the model selection process to identify why the live session model state is overwriting the fallback model choice (google/gemini-3.1-pro-preview) with the primary model (openai-codex/gpt-5.4).
• Consider adding a lock or a more robust state management system to prevent the live session model from overwriting the fallback model during execution attempts.
• Verify that the fallback model configuration is correctly applied and that the system can successfully execute the fallback model without interruptions from the live session model state.

Example

No specific code example can be provided without more details on the implementation, but ensuring that the fallback model is not overwritten by the live session state could involve adding checks or locks in the model selection logic.

Notes

The exact solution may depend on the internal workings of OpenClaw and how it manages model states and notifications. Understanding the flow of model selection and error handling will be crucial to resolving this issue.

Recommendation

Apply a workaround to prevent the live session model state from overwriting the fallback model choice, as this seems to be the primary cause of the failure to execute the fallback model successfully. This could involve temporary modifications to the model selection logic or the addition of a mechanism to lock the model choice once a fallback is triggered.