chatjimmy-proxy

OpenAI-compatible HTTP proxy for chatjimmy.ai. Point any OpenAI SDK or tool at it and use model jimmy.

Quick start

1. clone and install:

   </code></pre>
   </li>
   </ol>
   <p>git clone <repo>
   cd chatjimmy-proxy
   uv sync
   uv run playwright install chromium</p>
   <pre><code>2. configure:
      ```bash
   cp .env.example .env
   # edit PROXY_API_KEY (leave blank to disable auth)
   

   Note: if you have a PROXY_API_KEY set in your shell environment (for example some systems default it to your username), the proxy will require that exact value in the Authorization header. Use export PROXY_API_KEY= to clear it, or choose a different secret.

   3. run discovery once:

      </code></pre>
      </li>
      </ol>
      <p>uv run chatjimmy-discover</p>
      <pre><code>4. start proxy (default port 8000, change with `PORT` env var):
         ```bash
      uv run chatjimmy-proxy
      # or explicitly:
      uv run uvicorn chatjimmy_proxy.main:app --host 0.0.0.0 --port ${PORT:-8000}
      

      If you see "address already in use" set PORT to a free port (e.g. 8001) or kill the process currently listening on the port.

      Usage

      curl http://localhost:8000/v1/chat/completions \
        -H "Content-Type: application/json" \
        -H "Authorization: Bearer $PROXY_API_KEY" \
        -d '{"model":"jimmy","messages":[{"role":"user","content":"Hi"}]}'
      

      Streaming: add --no-buffer and "stream": true.

      Python example:

      from openai import OpenAI
      client = OpenAI(base_url="http://localhost:8000/v1", api_key="key")
      print(client.chat.completions.create(model="jimmy", messages=[{"role":"user","content":"Hi"}]).choices[0].message.content)
      

      Editors and agents

      Any tool that lets you supply a custom OpenAI‑compatible provider should work. You need three things:

      1. Base URL – the root of the OpenAI API, not a specific endpoint. Use http://<host>:<port>/v1 (omit /chat/completions); most clients append the path themselves. If you include /chat/completions twice you'll see 404s like /v1/chat/completions/chat/completions.
      2. API key – the secret from .env (or any string if auth is off).
      3. Model – jimmy.

      Correct Roo Code configuration example:

      {
        "provider": "OpenAI Compatible",
        "baseUrl": "http://localhost:8000/v1",
        "apiKey": "<your-proxy-key>",
        "model": "jimmy"
      }
      

      If you accidentally set the base URL to include /chat/completions, the agent will produce 404 errors when it tries to call /v1/chat/completions/chat/completions.

      Development

      uv run pytest tests/ -v
      uv run ruff check src/
      uv run ruff format src/
      

      Packaging & publishing

      Build with uv run hatch build. Releases are made by tagging vX.Y.Z; GitHub Actions tests and publishes to PyPI using PYPI_API_TOKEN.

      Troubleshooting

      • Port already in use – if the proxy fails to start with address already in use:

        # locate the offending PID
        sudo lsof -i :8000 -t   # substitute whatever port you were using
        # kill it (or choose a different port)
        sudo kill <pid>
        # or directly:
        sudo kill -9 $(sudo lsof -i :8000 -t)
        

        alternatively, set PORT to a free port before launching:

        PORT=8001 uv run chatjimmy-proxy
        

      • Discovery failures – run with HEADLESS=false or switch to mode: browser_relay in blueprint.

      • 401/403 – clear .jimmy_blueprint.json/.jimmy_state.json and re-run discovery; ensure PROXY_API_KEY is correct.

      • Slow first response – discovery runs on startup; subsequent requests are fast under HTTP‑replay mode.

      License

      MIT – see LICENSE.