liaocaoxuezhe383 downloadsSearch your vault semantically with local vector embeddings and expose RAG search to MCP clients.
Analogy adds local semantic search to Obsidian and exposes your indexed vault to MCP clients for retrieval-augmented generation.
The plugin is desktop-only because it uses Node.js APIs, starts a local ChromaDB process, and loads local embedding models.
127.0.0.1.Download the release assets from GitHub and copy them into:
<Vault>/.obsidian/plugins/analogy-rag-in-your-vault/
Required release assets:
main.jsmanifest.jsonstyles.cssEnable the plugin in Obsidian Settings -> Community plugins.
Obsidian only installs the three plugin assets above. Open Analogy settings and use the Local RAG runtime installer to install the embedding runtime in the plugin folder. Analogy also creates a small runtime package.json, installs the embedding runtime, and starts ChromaDB automatically when possible.
On first launch, the guided onboarding flow detects the local environment, downloads the pinned ChromaDB and embedding runtimes, verifies their checksums and signed release metadata, installs them atomically, and builds a small first index. Managed runtime downloads currently support macOS on Apple Silicon or Intel, and Windows on x64. The Ollama summary model is optional and can be skipped without blocking local semantic search.
Existing compatible Chroma indexes are migrated into the managed runtime without re-embedding. If migration or setup is interrupted, onboarding resumes from its saved recovery state; setup, repair, diagnostics, and rebuild controls remain available in Analogy settings.
If automatic installation fails, run the setup command inside the installed plugin folder:
npm run setup:local
If you need the companion MCP server files too, download the full runtime zip from the GitHub release.
Semantic Walk is a local, chunk-level exploration canvas in the main Obsidian workspace. Open it from the ribbon, the command palette, the current document, a sidebar search result, or a random indexed chunk.
Expanding a node queries the existing local index on demand. Multiple chunks from the same Markdown document can remain on the canvas while you follow another branch. Use New batch to reset the graph and choose a new random root. Relationship labels are relative to the current candidate batch rather than an absolute similarity percentage.
Indexes created by older Analogy versions may lack stable chunk identity, heading, or position metadata. Existing search remains compatible, but run Rebuild index to enable every Semantic Walk entry and obtain accurate chunk positioning.
127.0.0.1 by default.https://hf-mirror.com/.ANALOGY_ALLOWED_PATHS to restrict which vault paths the MCP server can return.npm install
npm run dev
Build production assets:
npm run build
Prepare release assets:
npm run release:prepare
The release tag must exactly match manifest.json version.
src/ - Obsidian plugin source.mcp-server/ - Companion MCP server source.scripts/ - Setup helpers such as model download..obsidian/ - Local development vault, ignored by the standalone plugin repo.Apache-2.0