Markdown Vector Database (mdvdb)

mdvdb indexes a folder of Markdown files and offers three search modes: hybrid search that fuses semantic embedding similarity with BM25 lexical search through reciprocal rank fusion, pure semantic search, and pure lexical search, all from a single command-line tool with everything stored on disk and no server or container to run.

Its documented use cases include hybrid search across documentation and runbooks with section-level results and frontmatter filters, an agent memory model that separates append-only daily logs from curated topic files connected through links, with recency-weighted decay and link-based re-ranking, documentation-site indexing with file watching and auto-clustering, drop-in retrieval for retrieval-augmented generation pipelines with JSON output, and non-destructive search over a personal Zettelkasten such as an Obsidian vault or Logseq graph. It is written in Rust.