Use a custom model with an assistant

Add the provider

Create or edit an assistant and open the model settings. Choose custom, enter the provider’s OpenAI-compatible base URL, model name, and API key.

The current assistant form may also request an embedding key for knowledge-base retrieval. Follow the labels shown for the selected assistant mode.

Test tools before production use

OpenAI-compatible text responses do not guarantee compatible tool calls, streaming, or structured output. Test chat first, then test every external action, search tool, phone flow, and knowledge-base path the assistant needs.

Use a provider and model you are authorized to call. Never paste a key into the assistant prompt or a customer conversation.