Summary
Docs you keep, rather than docs you re-fetch.
It builds a private index over the sites and repositories you actually rely on, so answers arrive with the source chunk attached instead of from memory. The queue and the repository watch are what keep an index built once from drifting out of date.
What it is
A documentation index you own. Point it at a URL or a local repository, let it fetch, chunk and embed the content, then search across everything it holds.
What you get
- Semantic search across everything indexed, with the source of each returned chunk
- A documentation site added by URL — fetching, chunking and indexing happen for you
- A local code repository indexed, with include and exclude patterns and chunking chosen by file type
- A watch on a repository, so the indexed copy tracks the working one
- The list of indexed sources and the list of indexed repositories with their settings
- A source or a repository removed when it goes stale, and a repository's configuration updated
- A processing queue you can list, run or clear, with an indexing status readout for the long jobs
- URLs extracted from text and checked against what is already indexed
Requirements
A vector store to write to and an embedding provider to reach it; a Docker Compose setup ships with the repository, and a web interface comes with it for managing the index outside the assistant. Configuration goes in the MCP client config.