What are VAPI's policies when it receives an error from a custom LLM? does it retry? I know we can set a timeout and if that is reached before the custom LLM streams to VAPI then the assistant won't retry the call, is the behavior the same when they get a 5XX error from the custom LLM provider?

Hi psicobloc, Yes, VAPI does implement retry logic when interacting with custom LLMs. You can configure this behavior using backoff plans. Specifically, for handling errors such as 5XX, you can use

Assistant.hooks.do[type=tool].tool.backoffPlan

. Here are the options available: - **Retry limits**: You can set

maxRetries

(0-10, default: 0). - **Timing adjustments**:

baseDelaySeconds

can be adjusted between 0-10 seconds. - **Exclusion settings**: You can configure to exclude specific HTTP status codes from retry attempts. When a request exceeds the configured

timeoutSeconds

, it will not be retried. You can set this duration between 20 and 600 seconds. This provides control over how long to wait for a response before timing out. [See more details on retry logic here](https://docs.vapi.ai/changelog/2025/8/27).