A knowledge base
that can answer back.
Connecting Wiki.js documentation to an AI retrieval layer, with a working keyword-search foundation and a semantic search index taking shape.
Separate service boundaries for the knowledge source and retrieval index.
Read project notes
The problem
Useful documentation needs to be retrievable from an assistant without giving the model unrestricted access to the systems behind it.
My approach
Connect Wiki.js through its GraphQL API and an MCP service. Keep the retrieval index separate, with header-aware chunking as a baseline and restricted tool access.
Working now
Keyword retrieval is working. PostgreSQL and pgvector are running, with vector similarity queries verified.
Next
Complete a small semantic-retrieval demo, then evaluate retrieval quality and indexing choices before scaling the document collection.