private-gpt

Why we included this project

Teams that want private AI without depending on cloud APIs usually end up rebuilding the same backend plumbing for every project. PrivateGPT is meant to remove that step. It does not run models itself; it is an API layer that connects your application to any OpenAI-compatible inference server and provides the higher-level building blocks: retrieval with citations, custom tools, MCP connectors, and structured access to databases and CSVs. Because the API follows the Claude API model, you keep a familiar interface while pointing it at your own model backend, and anything that speaks OpenAI-compatible endpoints can plug in. The bundled workbench UI is handy for demos and quick local testing, but the API is the real product. It also powers Zylon, an on-premise enterprise AI platform, which is a reasonable sign it holds up beyond a demo.