LINZA

A local-first sidecar that indexes Markdown folders into SQLite and exposes structured review workflows through MCP. It scans your Obsidian vault or docs directory, builds a map of material types and connections, then surfaces review intents you can accept or skip before anything writes. The agent_workspace tool gates all changes: doctor checks vault health, index builds the map, review_next shows proposed edits with evidence, and apply_review_items runs dry-run first. Embeddings are optional but unlock semantic search and topic clustering via LM Studio or Ollama. Designed for knowledge bases that have grown too large to navigate by memory, where you want the agent to suggest structure without autonomous writes.