YALP – Yet Another LLM Proxy

YALP is a tiny FastAPI based HTTP proxy that forwards any request to an up‑stream LLM service (e.g. OpenAI, Anthropic, etc.). It rewrites the reasoning_content field in JSON payloads to reasoning – a convenience used by some downstream tools – and streams the response back to the client unchanged otherwise.

Features

• Minimal, async implementation based on FastAPI, uvicorn, and httpx.
• Automatic environment‑variable configuration for host, port and upstream URL.
• Transparent proxying – all HTTP methods, query strings, headers (except host) and body are forwarded.
• Built‑in response transformer (ReasoningRenamer) that safely mutates JSON responses while preserving status codes and headers.

Installation

YALP can be installed from the source repository or from a built distribution. It requires Python 3.9+.

# From source (editable install)
pip install -e .

# Or a regular install
pip install .

The project also ships a uv.lock file; if you prefer uv you can install with:

uv pip install .

Configuration

Configuration is done via environment variables:

Variable	Description	Default
YALP_HOST	Interface address the proxy binds to.	127.0.0.1
YALP_PORT	TCP port the proxy listens on.	9333
YALP_BASE	Required. Base URL of the upstream LLM service (e.g. https://api.openai.com/v1).

The proxy will raise a RuntimeError at start‑up if YALP_BASE is not set.

Running the Proxy

After installing, a console script yalp is provided:

# Set the required upstream URL
export YALP_BASE="https://api.openai.com/v1"

# Optional custom host/port
export YALP_HOST="0.0.0.0"
export YALP_PORT="8000"

# Start the server
yalp

Quick request example

Once the proxy is running you can forward a request with curl (or any HTTP client). For instance, to proxy a chat completion request to an OpenAI‑style endpoint:

curl -X POST "http://$YALP_HOST:$YALP_PORT/v1/chat/completions" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
        "model": "gpt-4",
        "messages": [{"role": "user", "content": "Hello!"}]
      }'

The proxy will forward the request to YALP_BASE/v1/chat/completions, apply the ReasoningRenamer transformation (if applicable), and stream the response back to curl.

You can also run it directly with Python:

python -m yalp

The server will be reachable at http://<YALP_HOST>:<YALP_PORT>/ and will forward all paths to the upstream service.

Development

If you want to work on the codebase:

1. Clone the repository.
2. Install development dependencies (FastAPI, httpx, pytest, etc.) – they are already listed in pyproject.toml.
3. Run the proxy in a virtual environment as shown above.

Unit‑tests are not shipped yet, but you can add them under a tests/ folder and run them with pytest.

Contributing

Contributions are welcome! Please follow these steps:

1. Fork the repository.
2. Create a feature branch.
3. Make your changes, ensuring they keep the same minimal style as the existing code.
4. Open a pull request describing the change.

License

This project is licensed under the Apache License, Version 2.0. See the official license text at https://www.apache.org/licenses/LICENSE-2.0 for full details.