Codex (ChatGPT Subscription)

Provider value: codex

Use your ChatGPT Plus or Pro subscription to run ra agents instead of paying per-token API costs. This routes requests through OpenAI's Codex backend using your subscription credits.

TIP

OpenAI sanctions third-party tool usage via their "Codex for Open Source" program. Tools like OpenClaw and OpenCode use the same mechanism.

Setup

Step 1: Login

# Browser-based login (recommended)
ra login codex

# Headless/SSH environments
ra login codex --device-code

This opens a browser for OpenAI authentication. Tokens are saved to ~/.ra/codex-tokens.json.

Step 2: Run

ra --provider codex --model gpt-5.4 "Hello"

Authentication flows

Flow	Command	When to use
PKCE (default)	ra login codex	Desktop/laptop with a browser
Device Code	ra login codex --device-code	SSH, Docker, headless servers

Tokens refresh automatically. Re-run ra login codex if you get auth errors.

Models

Available models depend on your ChatGPT subscription tier:

Model	Plan
gpt-5.4	Plus, Pro
gpt-5.3-codex	Pro
o4-mini	Plus, Pro
o3	Pro

ra --provider codex --model o4-mini "Explain this error"

Environment variables

Variable	Required	Description
CODEX_ACCESS_TOKEN	No	Override the stored OAuth token (e.g. for CI). If not set, ra uses the token from ra login codex.

Config file

agent:
  provider: codex
  model: gpt-5.4

Or with an explicit token:

app:
  providers:
    codex:
      accessToken: ${CODEX_ACCESS_TOKEN}

agent:
  provider: codex
  model: gpt-5.4

How it works

1. ra login codex authenticates via OpenAI's OAuth (PKCE or device code flow)
2. The provider routes requests to chatgpt.com/backend-api/codex/responses — the same endpoint the official Codex CLI uses
3. Requests use the standard OpenAI Responses API format, so all ra features (tools, streaming, context) work normally
4. Usage draws from your ChatGPT subscription quota, not API billing

Limitations

• Subscription rate limits apply (30-150 msgs/5hr on Plus, 300-1,500 on Pro)
• Extended thinking (--thinking) is stripped from requests — the Codex backend may not support it
• Cannot use api.openai.com with subscription tokens — they only work with the Codex backend

See also

• OpenAI provider — for API key-based access
• Configuration — provider credentials reference