oaiproxy

Simple experimental proxy and local bridge that signs into a ChatGPT subscription account and exposes an OpenAI-compatible API on http://127.0.0.1:1456.

Current scope:

• POST /v1/chat/completions
• GET /v1/models
• GET /v1/models/:id
• GET /health
• POST /auth/login
• POST /auth/logout
• GET /auth/status
• GET /auth/callback

Requirements

• Node.js 22.11.0 or newer
• npm 10+
• A local browser session that can complete OpenAI OAuth

Install

npm install -g @nima__/oai-proxy

Run

oaiproxy

Default listen address:

• Host: 127.0.0.1
• Port: 1455

On startup:

• If a usable auth session is found, the server starts immediately.
• If auth is missing, the terminal prompts: No valid ChatGPT auth found. Launch browser login now? [Y/n]
• If you answer Y or press enter, the browser opens and the callback returns to http://localhost:1455/auth/callback

Run As A LaunchAgent

The clean macOS way to keep this running for a local OpenAI-compatible client is a user-scoped launchd agent.

That is exactly what I set up locally:

• a plist under ~/Library/LaunchAgents/
• absolute node path in ProgramArguments
• repo root as WorkingDirectory
• stdout/stderr sent to log files under ~/Library/Logs/oaiproxy/

Important: install auth first. The background agent has no interactive terminal, so do one successful login with:

oaiproxy

After auth exists, install the LaunchAgent:

oaiproxy-launchd install

Useful commands:

oaiproxy-launchd status
oaiproxy-launchd logs
oaiproxy-launchd uninstall

Defaults used by the LaunchAgent:

• Label: dev.oaiproxy.server
• Plist: ~/Library/LaunchAgents/dev.oaiproxy.server.plist
• Logs:
  • ~/Library/Logs/oaiproxy/stdout.log
  • ~/Library/Logs/oaiproxy/stderr.log

You can override the runtime parameters at install time:

PORT=1555 LOG_LEVEL=debug oaiproxy-launchd install

To change the port after the agent is already installed, edit the plist directly and reload:

# 1. Edit the PORT value in the plist
nano ~/Library/LaunchAgents/dev.oaiproxy.server.plist

# 2. Reload (this stops the running process and starts it on the new port)
launchctl unload ~/Library/LaunchAgents/dev.oaiproxy.server.plist
launchctl load  ~/Library/LaunchAgents/dev.oaiproxy.server.plist

Do not kill the process with kill or pkill — KeepAlive is set, so launchd will immediately restart it on the old port.

You can also pin a specific Node binary if your shell uses a version manager:

NODE_BINARY="$(command -v node)" oaiproxy-launchd install

If your reusable auth lives under a custom CODEX_HOME, pass that during install too:

CODEX_HOME="/path/to/codex-home" oaiproxy-launchd install

Configuration

Environment variables:

• HOST
  • Default: 127.0.0.1
  • Listen host for the local server.
• PORT
  • Default: 1455
  • Listen port for the local server.
• LOG_LEVEL
  • Default: info
  • Pino log level. Accepts fatal, error, warn, info, debug, trace, silent.
• UPSTREAM_TIMEOUT_MS
  • Default: 30000
  • Timeout used for OAuth token exchange and upstream Codex requests.

The OAuth callback URL is derived from the current PORT and always uses:

• http://localhost:<PORT>/auth/callback

Auth files

Primary auth file written by this project:

• ~/.chatgpt-codex/auth.json

Read-only fallback sources:

• $CODEX_HOME/auth.json
• ~/.codex/auth.json

If fallback auth is used, refreshes and new logins still write to ~/.chatgpt-codex/auth.json.

Endpoints

GET /health

• Simple liveness check.

GET /auth/status

• Shows whether auth is usable.
• Returns the source auth path, email, plan type, account id, and expiry.

POST /auth/login

• Starts OAuth login and opens the browser.

POST /auth/logout

• Removes the primary local auth file.

GET /v1/models

• Returns the tested model list.

GET /v1/models/:id

• Returns a single model record for supported model ids.

POST /v1/chat/completions

• Supports both stream: true and stream: false
• Converts OpenAI chat messages into the upstream Codex responses format
• Preserves multimodal user content with OpenAI-compatible image_url parts
• Translates upstream SSE back into OpenAI-compatible chat completion output

Quick checks

Health:

curl http://127.0.0.1:1455/health

Auth status:

curl http://127.0.0.1:1455/auth/status

Models:

curl http://127.0.0.1:1455/v1/models

Streaming chat completion:

curl -sN http://127.0.0.1:1455/v1/chat/completions \
  -H 'content-type: application/json' \
  -d '{
    "model": "gpt-5.4",
    "stream": true,
    "messages": [
      { "role": "system", "content": "Reply briefly." },
      { "role": "user", "content": "Say hello in three words." }
    ]
  }'

Non-streaming chat completion:

curl http://127.0.0.1:1455/v1/chat/completions \
  -H 'content-type: application/json' \
  -d '{
    "model": "gpt-5.4",
    "stream": false,
    "messages": [
      { "role": "system", "content": "Reply briefly." },
      { "role": "user", "content": "Say hello in three words." }
    ]
  }'

Multimodal chat completion:

curl http://127.0.0.1:1455/v1/chat/completions \
  -H 'content-type: application/json' \
  -d '{
    "model": "gpt-5.5",
    "stream": false,
    "messages": [
      { "role": "system", "content": "Describe images briefly." },
      {
        "role": "user",
        "content": [
          { "type": "text", "text": "What is important in this image?" },
          {
            "type": "image_url",
            "image_url": {
              "url": "data:image/png;base64,...",
              "detail": "low"
            }
          }
        ]
      }
    ]
  }'

Client setup

For any OpenAI-compatible client, use:

• Base URL: http://127.0.0.1:1455
• Model: gpt-5.4
• API key: leave blank unless the client insists on a value

The prototype route to use is /v1/chat/completions.

If you installed the LaunchAgent with a custom port, use that port in the client too.

Development

Typecheck:

npm run typecheck

Tests:

npm test