Smart Second Brain: an Obsidian plugin with semantic search, a topic graph and a note-reading agent

The vault-size problem Smart Second Brain is aimed at

Obsidian's built-in search matches strings. Once a vault passes a few thousand notes, you remember the idea of a note but not the words in it, and the backlink pane only shows links you made by hand. Smart Second Brain targets that gap from three directions at once: retrieval that mixes keyword matching with meaning, a graph that groups notes by topic rather than by explicit links, and an assistant that can read and write notes using both. The README frames it as "an AI assistant that actually knows your notes," and the plugin is aimed at people who already live in Obsidian rather than at teams evaluating a knowledge platform. The authors state they built it as a university project, that it is finished, and that they continue developing it in their spare time as "an experimental playground for state-of-the-art AI." That framing matters: this is a personal-vault tool with an experimental streak, not a product with a support contract.

How hybrid search and the Smart Graph are wired together

The README describes search as combining keyword matching with meaning, so the two retrieval paths run alongside each other and the result is the note you were thinking of rather than the one whose exact words you typed. The Smart Graph groups the vault by topic and draws connections that were never linked by hand, which means clustering runs over note content rather than over the link graph. Both features share the same substrate: an embedding model. The README is explicit that search and the graph work immediately with no configuration, and that adding an embedding model is what "unlocks semantic search and graph clustering." So the default install is a keyword-and-link experience, and the semantic layer is opt-in. The agent sits on top of both, with skills, memory, MCP support and integrations with other plugins, and the repository layout supports that reading: src/ holds the TypeScript sources, integration/ contains a test vault with the plugin symlinked into .obsidian/plugins/smart-second-brain, and the build is a Vite pipeline with Svelte components and Tailwind styling. The README does not document how embeddings are stored or which vector index is used, so anyone who needs to reason about index size or re-embedding cost has to read the source.

Installing Smart Second Brain from Obsidian's community plugins

The README gives one install path: install Smart Second Brain from Obsidian's community plugins and enable it. Search and the Smart Graph then work with no configuration. To go further you add an embedding model for semantic search and graph clustering, and an AI provider for the agent. The README points to the getting started guide on smartsecondbrain.dev for the details rather than repeating them in the repository, so the exact provider names and settings live on that site, not in the README.

For people building from source, the repository is a Bun and Vite project. The package.json defines the scripts. The development build watches and rebuilds on change:

bun run dev

A production build writes the plugin bundle:

bun run build

Tests run under Vitest, with a separate configuration for integration tests against a real vault:

bun run test
bun run test:integration

The integration setup script creates a test vault and symlinks the built plugin into it, which is how the repository exercises the plugin inside Obsidian rather than in a mocked environment:

bun run setup-vault

The README's install section does not cover manual installation into a vault's .obsidian/plugins directory, so treat the community plugin listing as the supported route and the source build as the contributor route.

What you give up: providers, index cost and an undocumented rollback

The first limitation is structural. Without an embedding model there is no semantic search and no graph clustering, and without an AI provider there is no agent. The README presents this as a graceful default, and it is, but it also means the headline features require a second decision about where your notes get processed. The README states that out of the box nothing leaves your machine, that the only network requests go to the providers you configure, and that there is no telemetry. That is a clear statement about the plugin's own behaviour, and it shifts the privacy question to whichever embedding model and AI provider you pick.

The second limitation is cost and maintenance of the index. Embedding a vault is not free at the provider level, and re-embedding after bulk edits is work the plugin has to do again. The README does not document how incremental re-embedding is scheduled, nor how to reset or rebuild the index if it drifts. The third is version coupling. A plugin that reads and writes notes and integrates with other plugins sits on Obsidian's plugin API; manifest.json and versions.json exist in the repository, but the README does not state a minimum Obsidian version, so check those files against the Obsidian release you run. Finally, the agent writes to your notes. The README describes agents that "read and write your notes" and shows a chat proposing an edit, but it does not document an undo path for those writes. If your vault is not under version control or synced with history, an agent edit is a real risk, and no part of the README addresses rollback.

Smart Second Brain compared with Obsidian's own search and a plain RAG script

The honest alternative is not another plugin. It is Obsidian's built-in search plus the core graph view, which cost nothing, need no provider and never send a note anywhere. The difference in approach is that built-in search is lexical and the core graph is built from links you created, while Smart Second Brain adds an embedding layer so retrieval can match meaning and the graph can cluster by topic. If your vault is small or heavily linked, the built-in tools already answer most queries and the plugin's value drops sharply.

The second alternative is a self-written retrieval script over the vault, typically a small pipeline that chunks Markdown, embeds it and queries it from a terminal. That gives you full control over chunking, storage and provider choice, and it keeps the index outside Obsidian. What it does not give you is the in-editor experience: a search pane, a graph rendered next to your notes, and an agent that can propose an edit inside the same window. Smart Second Brain's bet is that the integration is worth more than the control, and the repository's size (Svelte components, a Vite build, integration tests against a real vault) shows the maintenance that bet implies.

Licence, maintenance and upgrade cost

The plugin is MIT licensed, which permits commercial and private use, modification and redistribution provided the copyright notice and permission notice are retained. That is a permissive licence with no copyleft obligation on your own code, and it is the same licence Obsidian's own sample plugin uses. It says nothing about the terms of the embedding model or AI provider you connect, which are separate agreements you accept when you configure them; read those before pointing the plugin at a paid endpoint.

On maintenance, the repository is not archived. The last push was on 2026-09-05, and the most recent releases listed are 2.0.5, 2.0.4 and 2.0.3, all dated 2026-09-05. The package.json version is 2.1.1, ahead of the published 2.0.5, which is normal for a repository between releases but worth knowing if you build from source. The README describes the project as spare-time development, so plan for release cadence to be uneven. Upgrading a plugin that writes notes has two costs: the plugin bundle itself, which Obsidian manages, and any index you have built, which may need re-embedding after a change to chunking or the embedding model. Neither the README nor the release list documents index migration, so back up the vault before a major version jump.

Editorial conclusion

Adopt Smart Second Brain if you keep a large Obsidian vault and want meaning-based retrieval, an automatic topic graph and an agent that can edit notes, while staying on desktop or mobile with no telemetry. Do not adopt it if you need a hosted service with an SLA, or if you are unwilling to configure an embedding model and an AI provider, because the agent stays locked without them. Before relying on it, verify in the repository that manifest.json declares the minimum Obsidian version you run, that AGENTS.md matches the architecture of the version you install, and that your provider choice satisfies whatever data-handling rules apply to your notes.

Frequently asked questions

Official sources

1. License: MIT
2. Project website
3. README
4. Releases
5. s2b-dev/smart-second-brain on GitHub