Smart Second Brain: semantic search and note-writing agents inside an Obsidian vault
A free, open-source Obsidian plugin that makes your vault smarter: better search, an interactive knowledge graph, and an AI assistant that knows your notes.
At a glance
- What is it?
- Smart Second Brain is an MIT-licensed Obsidian plugin that layers keyword-plus-meaning search, a topic-clustered graph, and note-reading agents on top of a local vault. The interesting design choice is that search and the graph need no AI provider at all, while the agent does.
- Who is it for?
- Adopt Smart Second Brain if you already live in Obsidian, want semantic search and topic clustering without sending your vault anywhere, and accept that the agent half only turns on once you configure an embedding model and an AI provider. Skip it if you want a single binary that works offline end to end with no model setup, or if you need a documented schema for what the agent writes back into your notes.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository received new commits within the last day.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 16, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The gap Smart Second Brain targets in an Obsidian vault
Obsidian's built-in search matches strings. If you wrote a note about rate limiting under one phrasing and later search for it under another, you get nothing, and the backlinks panel only shows connections you remembered to create. Smart Second Brain attacks both problems from inside the vault. The README describes three pieces: search that combines keyword matching with meaning, a Smart Graph that groups the vault by topic and draws connections you never linked by hand, and agents that read and write notes. The intended user is someone with a vault large enough that manual linking has stopped scaling, who wants that help without shipping the vault to a hosted service by default. The README states plainly that out of the box nothing leaves the machine, the only network requests go to providers you configure yourself, and there is no telemetry.
What runs before you configure anything
This is the part most similar plugins get wrong, and the README is explicit about it: search and the graph work right away with no AI provider, and the install instructions say they work immediately with no configuration. Adding an embedding model is what unlocks semantic search and graph clustering. Adding an AI provider is what enables the agent. So the plugin degrades into a useful, fully local tool rather than a dead install waiting for an API key. That ordering matters for evaluation, because it means you can judge the search and graph on their own merits before deciding whether to involve a model at all. The README does not say which embedding models are supported or whether they run locally, only that you add one. That is the first thing to check in the getting started guide, since it decides whether semantic search stays on your machine or does not.
How the agent layer is put together
The repository carries an AGENTS.md file described as holding the architecture, commands and conventions, and a CONTRIBUTING.md with the development setup. From the README's feature list, the agent component includes skills, memory, MCP, and integrations with other Obsidian plugins, and the hero image caption describes a chat that reads your notes and proposes an edit. The word proposed is doing real work there: the agent suggests a change rather than silently rewriting files, at least in the flow the README illustrates. What the README does not give is the retrieval path. It never states how notes are chunked, how embeddings are stored, or whether the index persists across restarts. For a plugin whose headline claim is that the assistant knows your notes, that is a thin description, and the architecture document in the repository is where you would have to go to fill it in.
Installing it and the keys that matter
Installation is the standard Obsidian route: install Smart Second Brain from Obsidian's community plugins and enable it. The README gives no manual build steps, no BRAT instructions, and no settings keys by name. Configuration is described in prose rather than as a table: add an embedding model to unlock semantic search and graph clustering, add an AI provider to enable the agent. The plugin exposes no documented config file keys in the material available here, so anyone writing deployment notes for a team should treat the settings UI as the source of truth and not guess at key names. The homepage at smartsecondbrain.dev is where the getting started guide, the search and graph explainers, and the privacy page live. If you are evaluating this for more than one person, read the privacy page before the install page, because it determines what the provider configuration actually exposes.
Where the design gets thin
The README makes a broad privacy claim and then qualifies it in the same sentence: nothing leaves your machine except requests to providers you configure. Those two halves pull in opposite directions, and the README does not resolve which note content reaches a configured provider. If you add a hosted AI provider, the agent has to send something, and the documentation supplied here does not say what. The same gap applies to the embedding model. Semantic search over a vault requires embedding the vault, and whether that happens locally depends entirely on which model you pick, a choice the README leaves to the linked guide. There is also no statement about index size, rebuild time, or behaviour on very large vaults, and no migration notes for the 2.0.x releases, which shipped three patch versions in a single day according to the release timestamps. Rapid patch releases are not evidence of instability on their own, but they do mean you should read the changelog before upgrading a vault you depend on.
How it differs from wiring up a separate RAG stack
The obvious alternative is to keep Obsidian as a plain editor and build retrieval outside it: export notes to a folder, run a local embedding pipeline, and query it from a chat client. That approach gives you control over chunking, the vector store, and exactly which model sees which text, and it works across editors rather than only Obsidian. The cost is that you maintain the sync, the index, and the query layer yourself, and the graph view has no equivalent unless you build one. Smart Second Brain's bet is the opposite: keep everything in the plugin, accept the settings UI as the only interface, and get search and the graph working with zero setup. If your vault is the centre of your work and you do not want to run a pipeline, the plugin wins on effort. If you need to audit or customise the retrieval path, the external stack wins, because here that path is not documented in the README.
Maintenance, licence and upgrade cost
The project is MIT licensed, which permits commercial and private use, modification and redistribution provided the copyright notice and permission notice are preserved. That is a permissive licence, and nothing in it obliges you to publish changes. It also means no warranty, so the usual caution applies to relying on it for anything critical. On maintenance, the README is candid: the team built this initially as a university project, that phase is finished, and development continues in spare time as an experimental playground. That is a real signal about support expectations. You are adopting a plugin maintained by people with day jobs, with no stated release cadence and no stated compatibility policy for Obsidian versions. The practical upgrade cost is low for a single user, since it is a community plugin updated through Obsidian, but there is no documented rollback path and no index migration notes, so pinning a working version before a vault-wide reindex is the cautious move.
Editorial conclusion
Adopt Smart Second Brain if you already live in Obsidian, want semantic search and topic clustering without sending your vault anywhere, and accept that the agent half only turns on once you configure an embedding model and an AI provider. Skip it if you want a single binary that works offline end to end with no model setup, or if you need a documented schema for what the agent writes back into your notes. Before installing, read the privacy page at smartsecondbrain.dev/privacy/model/ and confirm exactly which providers receive note content, then check the getting started guide for the embedding model options, since that decision determines both search quality and what leaves your machine.
Community notes