LiteLLM

A LiteLLM proxy lets assist run Codex on any model the proxy serves, instead of the account you signed in to Codex with.

Connecting a proxy

Set the proxy’s URL and API key:

assist config set litellm.baseUrl https://litellm.example.com
assist config set litellm.apiKey sk-...

Then assist litellm list-models lists the models you can pick from.

Running Codex through the proxy

Set harness.codexModel to a model the proxy serves, and every Codex session runs on it:

assist config set harness.codexModel gpt-5.1-codex

Unset it to go back to Codex’s own provider. The Codex reviewer in a pull request review has its own key, review.codexModel.