Codex
OpenAI's Codex engine, connected via OAuth. Requires a ChatGPT Plus or Pro subscription. No API key — authentication is handled through a browser flow.
Setup
Authenticate from the TUI:
:keys set openai connect
This opens a browser window. Once you approve, Codex models appear in the model palette. The connection auto-refreshes — you won't need to reconnect unless you explicitly disconnect.
To manage the connection:
:keys set openai reconnect # re-authenticate
:keys set openai disconnect # remove authentication
:keys show # check current state
Configuration
import { Codex } from "@axon/engines"
export default defineAgent({
engine: Codex(),
})
With options:
import { Codex } from "@axon/engines"
export default defineAgent({
engine: Codex({
model: "gpt-5.5", // default
effort: "medium", // low | medium | high | xhigh
}),
})
effort controls reasoning depth on models that support it. Higher effort increases
latency and token usage. Omit it for non-reasoning models.
Switching models
Open the model palette with *. All Codex models appear once authenticated. The
selected model overrides the model field in config for the current session.