Graph Mind

Graph Mind turns your vault into a local, private knowledge search and chat experience. It indexes your Markdown notes, ranks relevant passages, and lets you ask questions in a dedicated chat view with citations to your notes.

What this plugin does

• Builds a local index of your vault’s Markdown notes.
• Supports advanced search filters (tags, path, extensions, exact phrases).
• Reranks results with local embeddings for better relevance.
• Lets you chat with your vault in a side panel and jump to cited notes.
• Runs locally; no cloud dependency is required.

Requirements

To use chat and semantic search, you need a local Ollama server running on your machine with at least:

• One chat model (for answers).
• One embedding model (for semantic search).

If Ollama is not running, the plugin can still open the UI but it won’t be able to generate answers or embeddings.

Install

Manual install:

1. Copy main.js, manifest.json, and styles.css into: VaultFolder/.obsidian/plugins/graph-mind/
2. Reload Obsidian.
3. Enable the plugin in Settings → Community plugins.

Quick start

1. Open Settings → Graph Mind.
2. Set the Ollama base URL (default is http://localhost:11434).
3. Pick your LLM model and Embedding model.
4. Use the ribbon icon (bot) or the command Open Graph Mind chat to open the chat view.
5. Ask a question and review the cited sources.

User guide

Commands

• Build Graph Mind index: Rebuilds the full vault index.
• Open Graph Mind chat: Opens the chat view in the right sidebar.

Chat view

• Ask questions in natural language.
• The plugin first searches your vault, then generates an answer based on top results.
• Citations appear as numbered links that open the source notes.

Advanced search syntax

You can use filters inside questions to refine search:

• Exact phrase: "project kickoff"
• Tag: #meeting
• File extension: ext:md
• Path include: path:Projects/2025
• Path exclude: -path:Archive
• Exclude word: -draft

Example:

"design review" #ux path:Projects/2025 -draft

Tips

• Use Build Graph Mind index after major changes or large imports.
• If results feel off, check your embedding model in settings.
• Keep short, focused notes for better search precision.

Privacy

Your notes stay on your device. The plugin indexes locally and only communicates with the Ollama server you configure in settings (the Ollama base URL) to generate chat responses and embeddings.

License

GNU General Public License v3.0 (GPL-3.0-only). See LICENSE.

Third-party licenses

This plugin includes the following third-party libraries (license as reported by their package metadata):

• @mastra/rag — Apache-2.0
• @xenova/transformers — Apache-2.0
• minisearch — MIT
• nodejieba — MIT
• ollama — MIT
• onnxruntime-web — MIT
• path-browserify — MIT
• voy-search — MIT OR Apache-2.0