Grok Build CLI: Models and Custom Endpoints

@GrokInsider· Jun 6, 2026

Listen0:00 / 2:11

Grok Build defaults to xAI's models, but it is not locked to them. You can switch models mid-session, set a project default, or point Grok at Anthropic, OpenAI, Ollama, or any OpenAI-compatible endpoint. This part shows how.

First time here? Start with Grok Build CLI: Zero to Hero (Start Here).

Pick a model

List everything available, choose one per run with -m, or switch live with /model (alias /m) or the Ctrl+M picker. The default for new sessions is grok-build.

Add a custom model

Custom endpoints live under [model.<name>] in ~/.grok/config.toml. Grok speaks three API backends, selected with api_backend: OpenAI Chat Completions (the default), OpenAI Responses, and Anthropic Messages. Pair each with the right auth_scheme, bearer for most providers, x_api_key for Anthropic.

Run models locally

Point Grok at a local Ollama server and your code never leaves the machine. Any OpenAI-compatible server works the same way, just change the base_url.

Override a built-in, keep the defaults

You do not have to redefine a model to tweak it. Use the model's name as the section key and set only the fields you want to change. Grok starts from the built-in defaults, including the correct base_url, and applies your overrides on top. Credential resolution runs api_key, then env_key, then the global XAI_API_KEY fallback.

The web search model is separate

The web_search tool uses its own model, configured under [models]. If you point it at a custom model, that model must use the Responses API backend. The context_window value you set is what tells Grok when to auto-compact, so set it accurately for custom endpoints.

For enterprise setups, combine a custom base_url proxy with an external auth provider and auto_update = false for a fully managed deployment.

Continue the series

The full series