What it is
An open-source indexing layer that keeps the vault as plain files while generating a disposable local database outside the vault and outside the repo.
Project page
A local read model for Markdown vaults.
VaultLayer turns a Markdown or Obsidian vault into a rebuildable local database for full-text search, vector retrieval, WikiLinks, metadata, provenance, CLI, and MCP access.
markdown obsidian sqlite vectors mcp
Local-first
runtime model
Provenance
retrieval rule
CLI + MCP
interfaces
Project notes
Why it exists
Agents should not repeatedly crawl large Markdown folders. They need bounded, cited retrieval with source path, heading, hash, timestamp, and excerpt.
Useful for
Local vault search, retrieval benchmarks, vector experiments, MCP tool access, and replacing one-off indexing logic in vault-facing apps.
How it works
The indexer reads Markdown or Obsidian files, builds SQLite/FTS and vector-ready projections, and exposes search results with provenance.
Safety model
Private vault text, generated databases, embeddings, caches, and benchmark outputs stay outside the repository.
Current focus
Pilot-ready local MVP: package smoke tests, doctor checks, benchmark scripts, synthetic messy-vault preflight, and explicit remote-sync boundaries.
Why it matters
It separates storage from retrieval so viewers, agents, and MCP tools can share one rebuildable vault query layer.
Next action
Open GitHub for the current CLI commands, backend details, benchmarks, and implementation state.
This page is the human-readable project entrypoint. GitHub remains the place for current files, implementation details, issues, and release state.