model-switch

To change the Manager Agent's LLM model, you run the provided model-switch script with the desired model id, which verifies reachability via the AI Gateway and applies a persistent configuration update. If the script outputs RESTART_REQUIRED, you must restart the gateway to apply the change.

Yes, you can set context window size and toggle reasoning mode when switching the LLM model by passing optional flags during the update script execution. For unknown model ids, you must provide an explicit override for context window and expected max tokens.

If the target model is unreachable through the AI Gateway, the model switch will stop and return a clear error. This pre-check prevents breaking runtime connectivity and ensures the system is not left in an inconsistent configuration.

Yes, you need bash, curl, and jq installed in your environment to run the model-switch script, as these dependencies are required to test gateway reachability and update the persistent configuration files.

If changing the default provider fails during an LLM model switch, you should create a new AI Provider and AI Route instead of modifying the default provider. This operational safety measure prevents breaking existing gateway connectivity.