Codex with ChatGPT login¶

Use the codex provider with a Codex ChatGPT account login. Chat and streaming use the Responses endpoint.

Get started¶

Install Republic using the quickstart. With an existing Codex file login, replace MODEL_ID with a model available to your account and run:

import asyncio

import republic


async def main():
    model = republic.get_model("codex:MODEL_ID")
    response = await model.chat("Say hello in one sentence.")
    print(response.text)


asyncio.run(main())

Authentication¶

The default credential file is $CODEX_HOME/auth.json, or ~/.codex/auth.json when CODEX_HOME is unset. Credentials must be available in that file; a CLI login stored only in the keyring cannot be read through this path.

To choose another file, replace the model construction with get_model(..., auth=CodexAuth.from_file(path)). Import CodexAuth from republic.providers.

Log in explicitly¶

Inside an async function, obtain an auth object and pass it to the model:

import republic
from republic.providers import CodexAuth

auth = await CodexAuth.login()
model = republic.get_model("codex:MODEL_ID", auth=auth)

This runs the installed Codex CLI and selects file storage for this invocation. Use device_auth=True for device authorization or executable= to select the CLI executable. Model requests do not start this login flow themselves.

Store and refresh credentials¶

CodexAuth.from_file() reads the file immediately and before every request. Expiring tokens are refreshed when a refresh token is available, then saved to the same file.

For application-managed credentials, use CodexAuth(token, account_id=...). Reuse the auth object across requests and persist its updated auth.token in your own credential store.

Model support¶

Use chat() for a complete response or stream() for events. Responses is the only supported format; max_tokens is unsupported. Model access depends on your ChatGPT account.

For OpenAI API-key access, use the OpenAI provider. See configuration for credential precedence and HTTP clients.