obsidian-semantic-memory

Semantic memory for Obsidian — local-first indexing, semantic retrieval, FTS, entities, facts, and MCP tools for agents.

[](https://github.com/sterben-enec/obsidian-semantic-memory/actions/workflows/ci.yml) [](https://nodejs.org/) [](./LICENSE)

[Русская версия → README.ru.md](README.ru.md)

obsidian-semantic-memory turns an Obsidian vault into a local memory layer that agents and tools can query.

It indexes Markdown notes into SQLite + vector search and exposes a retrieval stack that combines:

• semantic similarity
• BM25 full-text search
• entity lookup
• graph relations
• structured facts
• recency-aware ranking

It is built for local-first personal knowledge systems — not cloud dashboards, not hosted SaaS search.

• Private by default — local embeddings, local DB, localhost API
• Better than plain vector search — semantic + FTS + graph + facts
• Agent-friendly — MCP tools for OpenClaw, Claude Code, and other MCP clients
• Obsidian-native — works directly on Markdown notes and wikilinks
• Live updates — watcher reindexes changes in seconds

• Local-first embeddings via @xenova/transformers
• Multi-signal retrieval with semantic vectors + BM25 FTS + graph/fact boosts
• MCP server with 7 tools
• HTTP API for local integrations
• Structured memory with entities, relations, and facts
• `.semanticignore` support for excluding private/noisy paths

git clone https://github.com/sterben-enec/obsidian-semantic-memory
cd obsidian-semantic-memory
cp .env.example .env
npm install
npm run build

Set VAULT_PATH in .env, then:

# full rebuild
node dist/cli.js reb