CodexOAuth

Direct ChatGPT Codex OAuth access from a local CLI and an OpenAI-compatible local API.

It sends requests straight to the ChatGPT Codex backend:

your app
-> local CodexOAuth CLI/API
-> https://chatgpt.com/backend-api/codex/responses
-> GPT-5.4

It sends structured Responses requests directly and does not require an OpenAI API key. It uses a Codex OAuth access token as a bearer token for the ChatGPT Codex backend.

What Is Included

• CodexOAuth: standalone CLI for sending one request directly to the Codex backend.
• codex_oauth_api.py: local FastAPI server exposing OpenAI-style /v1/responses and /v1/chat/completions.
• docs/SETUP.md: setup and usage guide.
• docs/HOW_IT_WORKS.md: explanation of the OAuth/backend flow.
• docs/TOOL_CALLING.md: function/tool-calling compatibility notes.

Install

git clone https://github.com/NavilanSanthanakrishnan/CodexOAuth.git
cd CodexOAuth
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt

Auth

Fastest path if you already use Codex CLI:

./CodexOAuth --status

If your ~/.codex/auth.json token is valid, this will work immediately.

Recommended path for a separate app session:

./CodexOAuth --login

That runs the OpenAI Codex device login flow and stores private tokens in:

~/.codex-oauth/auth.json

Using a private store avoids refresh-token conflicts with Codex CLI or editor extensions.

CLI Usage

./CodexOAuth "Write one sentence about espresso."

Use a different reasoning effort:

./CodexOAuth --reasoning medium "Explain OAuth in two bullets."

Return JSON:

./CodexOAuth --json --no-stream "Reply with exactly: ok"

API Usage

Start the local API:

source .venv/bin/activate
python codex_oauth_api.py --host 127.0.0.1 --port 8766

Use this OpenAI-compatible base URL:

http://127.0.0.1:8766/v1

For OpenAI-compatible clients:

Base URL: http://127.0.0.1:8766/v1
API key: anything
Model: gpt-5.4

Responses API:

curl http://127.0.0.1:8766/v1/responses \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "gpt-5.4",
    "input": "Write one sentence about espresso."
  }'

Chat Completions:

curl http://127.0.0.1:8766/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "gpt-5.4",
    "messages": [
      {"role": "user", "content": "Write one sentence about espresso."}
    ]
  }'

Python OpenAI SDK:

from openai import OpenAI

client = OpenAI(
    base_url="http://127.0.0.1:8766/v1",
    api_key="anything",
)

response = client.responses.create(
    model="gpt-5.4",
    input="Write one sentence about espresso.",
)

print(response.output_text)

Local API Auth

By default the local API accepts local requests without a bearer token. To require one:

export CODEX_OAUTH_API_KEY="choose-a-local-secret"
python codex_oauth_api.py --host 127.0.0.1 --port 8766

Then requests need:

-H "Authorization: Bearer choose-a-local-secret"

More

Read docs/SETUP.md for setup details, docs/HOW_IT_WORKS.md for the backend flow, and docs/TOOL_CALLING.md for tool-calling behavior.

For RLM integration notes, read docs/RLM.md.