Antigravity OpenAI-Compatible Proxy

A small FastAPI wrapper that accepts OpenAI-compatible requests and invokes the local agy CLI. For a chat request selecting claude-sonnet-4-6, it runs agy --model claude-sonnet-4-6 --prompt "...". /v1/models runs agy models and exposes the returned model IDs. The CLI owns sign-in/OAuth; model names are passed through unchanged.

Run on Termux (Android)

Install Python and unzip in Termux if needed. The agy command must already be installed, available in Termux's PATH, and signed in:

pkg update
pkg install python unzip

Unzip the project, enter its folder, install the package and dependencies, then start it:

unzip antigravity-proxy.zip
cd antigravity-proxy
cp .env.example .env
python -m pip install .
antigravity-proxy

The defaults are AGY_EXECUTABLE=agy and PROXY_PORT=8000. Change these in .env if needed. For a local development run instead of installing, use python main.py.

Test the proxy:

curl http://127.0.0.1:8000/health
curl http://127.0.0.1:8000/v1/models
curl -N http://127.0.0.1:8000/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model":"YOUR_MODEL","messages":[{"role":"user","content":"Hello"}],"stream":true}'

/health runs agy models and returns HTTP 200 if the command succeeds, or 503 if it cannot start or returns an error. Chat requests execute agy --model <model> --prompt <prompt> without shell interpolation. Streaming requests are returned as SSE while stdout is produced by the CLI.

Build and upload to PyPI from Termux

PyPI does not accept a project-source ZIP as a Twine upload. Twine uploads the source distribution and wheel built from this project. From the unzipped project directory:

python -m pip install --upgrade build twine
python -m build
python -m twine upload dist/*

Twine will prompt for credentials. For a PyPI API token, enter __token__ as the username and the complete token (including the pypi- prefix) as the password. Create the project/token on PyPI first. The name antigravity-proxy must be available on PyPI; if it is already taken, change project.name in pyproject.toml to an available name and rebuild before uploading.