TypeScript port of a Codex/ChatGPT account load balancer with a dashboard, SQLite storage, and OAuth-backed account onboarding.

This repo currently provides:

• Fastify backend on 2455
• React dashboard on 5173
• OAuth callback listener on 1455
• SQLite persistence for accounts, settings, and request logs
• OAuth account storage with refresh-token based renewal

• Start backend and frontend together with one command
• Add accounts from the dashboard using real OpenAI OAuth
• Configure Codex CLI to point at this proxy
• Persist and refresh OAuth account tokens

OAuth account onboarding is implemented, but full codex-cli request-shape compatibility is still incomplete.

The upstream ChatGPT Codex backend is stricter than standard OpenAI Responses API. Until the request adapter is completed, direct proxy requests may still require upstream-native fields like:

• instructions
• input as a list
• store: false
• stream: true

Install dependencies:

npm install
cd frontend && npm install --legacy-peer-deps && cd ..

Run backend and frontend together:

npm run dev

Services:

• backend API: http://127.0.0.1:2455
• frontend dashboard: http://127.0.0.1:5173
• OAuth callback: http://127.0.0.1:1455/auth/callback

Backend only:

npm run dev:backend

Frontend only:

npm run dev:frontend

Run the full project in Docker:

docker compose up --build

Services:

• backend API: http://127.0.0.1:2455
• frontend dashboard: http://127.0.0.1:5173
• OAuth callback: http://127.0.0.1:1455/auth/callback

The compose setup exposes port 1455 so browser OAuth can finish outside the container.

Defaults are already set in code, but these are the main ones:

HOST=0.0.0.0
PORT=2455
DB_PATH=./data/codex-lm-ts.db

AUTH_BASE_URL=https://auth.openai.com
OAUTH_CLIENT_ID=app_EMoamEEZ73f0CkXaXp7hrann
OAUTH_SCOPE="openid profile email offline_access"
OAUTH_REDIRECT_URI=http://localhost:1455/auth/callback
OAUTH_CALLBACK_HOST=127.0.0.1
OAUTH_CALLBACK_PORT=1455

For Docker, the compose file already overrides OAUTH_CALLBACK_HOST=0.0.0.0.

1. Start the project.
2. Open http://127.0.0.1:5173.
3. Go to Accounts.
4. Click Add with OAuth.
5. Choose:
   • Browser (PKCE) for normal local login
   • Device code for headless or remote setups
6. Complete the OpenAI sign-in flow.

After success, the account is stored in SQLite and will be refreshed automatically before proxy use.

• The callback listener uses http://localhost:1455/auth/callback
• Port 1455 must be reachable from your browser
• If the callback cannot reach the app directly, paste the callback URL into the dialog’s manual callback input

The dashboard still supports manual upstream accounts, but if your goal is Codex/ChatGPT OAuth-backed usage, use the OAuth dialog instead.

Edit ~/.codex/config.toml:

model = "gpt-5.3-codex"
model_provider = "codex-lm-ts"

[model_providers.codex-lm-ts]
name = "OpenAI"
base_url = "http://127.0.0.1:2455/backend-api/codex"
wire_api = "responses"

If you enabled proxy API key enforcement in dashboard settings, add:

[model_providers.codex-lm-ts]
name = "OpenAI"
base_url = "http://127.0.0.1:2455/backend-api/codex"
wire_api = "responses"
env_key = "CODEX_LM_TS_API_KEY"

And export it:

export CODEX_LM_TS_API_KEY="your-proxy-key"

Backend health:

curl http://127.0.0.1:2455/health

List models through the proxy:

curl http://127.0.0.1:2455/backend-api/codex/models

Test a manual upstream-native request:

curl -N -X POST http://127.0.0.1:2455/backend-api/codex/responses \
  -H 'content-type: application/json' \
  -d '{
    "model": "gpt-5.3-codex",
    "instructions": "You are a helpful coding assistant.",
    "store": false,
    "stream": true,
    "input": [
      {
        "role": "user",
        "content": [
          {
            "type": "input_text",
            "text": "Write a hello world program in TypeScript."
          }
        ]
      }
    ]
  }'

The proxy supports Cherry Studio via both the OpenAI Responses API and Chat Completions.

1. In Cherry Studio, go to Settings → Providers and add a new provider.
2. Choose OpenAI (Responses API) — not "OpenAI-Compatible".
3. Set API Base URL to http://localhost:2455 (or your backend URL). Do not add /v1 — Cherry Studio appends it.
4. Set API Key to your proxy API key (from dashboard Settings) if you enabled proxy auth.
5. Add the Codex models (e.g. gpt-5.3-codex) and enable the provider.

Cherry Studio will send requests to POST /v1/responses, which the proxy forwards to Codex. The Responses API format is compatible with Codex.

If you prefer OpenAI-Compatible (Chat Completions), use the same base URL. The proxy adapts requests to Codex and transcodes the streamed response.

• GET /health
• GET /api/accounts
• POST /api/accounts
• PATCH /api/accounts/:id
• POST /api/oauth/start
• GET /api/oauth/status
• POST /api/oauth/complete
• POST /api/oauth/manual-callback
• GET /api/settings
• PUT /api/settings
• GET /api/dashboard/summary
• GET /api/request-logs
• GET /v1/models
• POST /v1/chat/completions (Cherry Studio / OpenAI Chat Completions)
• POST /v1/responses
• GET /backend-api/codex/models
• POST /backend-api/codex/responses

Backend:

npm run typecheck
npm test

Frontend:

cd frontend
npm run typecheck
npm run build