Minerva

In early 2025, I got this idea with a friend of mine, Paul Xue, that RAG is huge and very useful for AI-powered apps, but so many of the existing solutions are hosted. Meaning that if you are within a company where you have proprietary data or various PII that you would like to only be hosted on premise you have a problem. You can’t use the existing RAG solutions.

So we decided to build Minerva (the goddess of knowledge). Minerva is a self-hosted API that turns a bucket of documents into a searchable RAG knowledge base. Point it at object storage and it handles the unglamorous middle:

1. pull the documents
2. extract and semantically chunk the text
3. embed the chunks
4. serve retrieval

We tried to sell it to companies but didn’t find as much appetite as we were hoping. The initial prototype still exists tho. If your machine can host Docker images, you can try it out.