Skip to main content
Grok Build is xAI’s official CLI assistant. It supports custom OpenAI-compatible endpoints via ~/.grok/config.toml, so any model in the AI Gateway catalog can be reached from a single config entry without changing tooling.

​
Prerequisites

​
Install Grok Build

The installer places grok and agent in ~/.grok/bin/ and updates the user PATH.

​
Setup

Always use the provider/model_id format for model IDs (e.g. anthropic/claude-sonnet-4-6, not claude-sonnet-4-6). Bare IDs return a 400 invalid model format error.
1

Edit ~/.grok/config.toml

Configure models with a [models] section and one [model.<key>] block per model. The example below uses Claude Sonnet 4.6 routed through Orq.ai:
Replace anthropic/claude-sonnet-4-6 with any model identifier from the AI Gateway catalog. To register additional models, add more [model.<key>] blocks and switch between them with -m <key>.
2

Export the API key

Replace <orq-api-key> with the key from Organization > API Keys. The env_key field in config.toml tells Grok Build which environment variable to read.
3

Launch Grok Build

Start an interactive session with the Orq-routed model:
For a headless one-shot completion:

​
Configuration Reference

​
config.toml fields

​
Adding more catalog models

Add one [model.<key>] block per model and use -m <key> to select at runtime:

​
Troubleshooting

If the npm package @vibe-kit/grok-cli is installed, the bare grok command may resolve to it instead of the xAI binary. Use the absolute path ~/.grok/bin/grok or remove the npm package to restore the correct resolution.
Confirm the environment variable named in env_key is exported in the current shell and contains a valid Orq.ai API key. Run echo $ORQ_API_KEY to verify.
Confirm base_url is set to https://api.orq.ai/v3/router and env_key points to a valid Orq.ai API key (not an xAI or OpenAI key). Requests to the wrong endpoint will not appear in Orq.ai Traces.
Verify the model field uses the exact AI Gateway identifier (e.g. anthropic/claude-sonnet-4-6). Check the available models in Workspace Settings → Models.

​
Verification

Launch Grok Build with grok -m orq-claude and send a prompt. The response appears in the session and the trace appears in Orq.ai Traces with the model identifier anthropic/claude-sonnet-4-6 (or whichever model was selected).