mlt-oss/open-assistant-api ? reverse-engineered prompt

Build me a self hosted AI assistant API that feels like the OpenAI assistant endpoints, so I can plug it into the normal OpenAI Python client and start creating assistants right away.

It should run locally with Docker Compose, expose a clean API and docs page, and let me create assistants, chat with them, and stream responses. I want it to support different model providers through a simple config, plus basic RAG for uploaded files and search style retrieval. Please include support for tools and custom functions too, so assistants can call external actions like web search or other services. Add simple token based auth for isolating users, and make sure there is a clear way to configure model keys and optional search keys from environment variables.

If you can, wire in a small playground or example so I can try it fast. Use current docs online if you need to.