2025-12-04

The vLLM and llama.cpp shape, without a key

Those servers speak a small OpenAI HTTP surface. This host serves that surface on two ports.

vLLM and llama.cpp both grew an OpenAI-compatible HTTP server so existing clients could change a base URL and keep their code. The routes that matter are GET /v1/models, POST /v1/chat/completions, and POST /v1/completions. Embeddings, when present, are POST /v1/embeddings. Health is often GET /health or GET /v1/health.

Authentication is a flag, not part of the protocol. With the flag off, the server accepts a request that has no Authorization header. The model string in the body is whatever the client sends. The server on this host uses gpt-4o-mini when the field is missing.

A client written for either project can point at this host without a local GPU. Use port 8000 or port 8080. They implement the same routes.

On this host

No API key. The chat route on port 8000 is the one to try first. Port 8080 accepts the same body.

curl http://datattention.com:8000/v1/models
curl http://datattention.com:8000/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model":"gpt-4o-mini","messages":[{"role":"user","content":"Hello"}]}'

The full list of ports and paths is on the endpoints page.