simpleopenai-proxy

A lightweight transparent proxy for OpenAI-compatible Chat Completions endpoints.

It sits in front of a non-standard upstream API and exposes a standard local OpenAI Chat Completions server. Currently it normalizes cases where upstream providers incorrectly emit empty strings ("") for content or reasoning_content instead of the expected OpenAI null.

Install / run with uvx

uvx simpleopenai-proxy <source-api-address> <localport>

Example:

uvx simpleopenai-proxy https://api.example.com/v1/chat/completions 8080

The proxy transparently forwards all request headers (including Authorization) to the upstream endpoint, so clients use the local server exactly like a standard OpenAI endpoint:

curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <upstream-api-key>" \
  -d '{
    "model": "glm_5_dot_2_public",
    "messages": [{"role": "user", "content": "hello"}],
    "stream": true
  }'

Run from source

uv run server.py <source-api-address> <localport>

Current normalizations

• choices[].message.content → null when upstream sends ""
• choices[].message.reasoning_content → null when upstream sends ""
• choices[].delta.content → null when upstream sends "" (streaming)
• choices[].delta.reasoning_content → null when upstream sends "" (streaming)
• SSE lines without a space after data: are parsed and re-emitted as standard data: {...}

Development

uv run python -m unittest discover tests