RushDB

Index your vault into RushDB and turn your notes into a graph you and your AI agents can search and traverse.

• Sync: notes, properties (frontmatter), tags, headings and [[links]] become RushDB records and LINKS_TO relationships. Changes, renames and deletions follow in seconds.
• Search: keyword (BM25) search over your notes, or hybrid keyword + semantic search with embeddings turned on.
• Related notes: a sidebar for the open note with its links, backlinks, notes two links away, similar notes and notes that share its tags.
• Agent access: point the RushDB MCP server at the same project and any MCP client can query your vault: filter by properties, follow links, search by meaning.

Your Markdown files stay the source of truth. The plugin only ever writes to RushDB, never to your vault.

Setup

1. Install RushDB from Settings → Community plugins → Browse, then enable it. It needs Obsidian 1.13 or later.
2. Choose where your notes go, in Settings → RushDB:
   • RushDB Cloud: create a project at app.rushdb.com, create an API key and paste it under API key; Obsidian keeps it in its secret storage (as rushdb-api-default unless you pick another secret). Keep the API URL as https://api.rushdb.com.
   • Self-hosted platform: set the API URL to your own RushDB platform and use one of its API keys.
   • Standalone engine: run rushdb serve --port 3999 and choose Standalone engine. Add --data-dir <path> to keep the data across restarts.
3. Folders to index starts as /, the whole vault. Narrow it to the folders to send, and list any folders to skip. Nothing is sent until RushDB is connected.
4. Close the settings. The first sync starts, and the status bar shows its progress.

Open the search with the RushDB: Search notes command and the sidebar with RushDB: Show related notes.

Semantic search

With Semantic search on, RushDB embeds the text of each note with its managed embedding model. Search then ranks by meaning as well as keywords, and the sidebar finds similar notes by meaning. This needs RushDB Cloud, or a self-hosted platform configured with an embedding model. A standalone engine has no embedding model, so it uses keyword search only.

Connect an AI agent

Add the RushDB MCP server to your agent with the same API key:

{
  "mcpServers": {
    "rushdb": {
      "command": "npx",
      "args": ["@rushdb/mcp-server"],
      "env": { "RUSHDB_API_KEY": "<your API key>" }
    }
  }
}

The agent then sees your notes as NOTE records with path, title, folder, tags, headings, body, modifiedAt, linkKey (<vault id>:<path>), linksTo (the linkKeys of the indexed notes it links to) and your properties (prefixed fm_, e.g. status → fm_status), joined by LINKS_TO relationships (with kind: "wikilink"). Try: "Which notes tagged project link to notes that are still fm_status: draft?"

What is sent, and where

This plugin uses the network. It is clear about what it sends:

• What: for each note in the folders you chose, its path, title, tags, aliases, headings, properties, modification time and text (up to the limit in Text per note), plus the links between those notes. Notes outside those folders, attachments and other files are never sent, and the plugin only walks the folders you chose: it does not list the rest of the vault.
• Where: only to the server you configure: RushDB Cloud, your own RushDB platform or your own engine.
• Your data: on RushDB Cloud it is stored in your project and used to answer your queries. With semantic search on, note text is embedded by the RushDB platform. See the privacy policy.
• Removal: Remove this vault from RushDB in the settings deletes everything this vault sent.
• Credentials: the API key and engine token are kept in Obsidian's secret storage, not in the plugin's data file. Keys saved by versions before 0.3 are moved there automatically.

No telemetry, no analytics and no ads. The plugin's source is in this repository, unminified in development builds.

Account and payments: RushDB Cloud requires an account and has a free tier and paid plans. The plugin is free, and it works fully with a self-hosted platform or a standalone engine without any account.

Development

See CONTRIBUTING.md.

License

Apache-2.0