2.0.0

The first stable release since 1.3.0. Smart Second Brain has grown from a chat-with-your-notes plugin into three features that each work on their own: search, a graph that groups your notes into topics, and an agent that can read and edit your vault with your approval.

Full feature documentation is at smartsecondbrain.dev.

🔍 Search

• Full-text search with fuzzy matching, tag and folder filter chips, and recency-aware ranking.
• PDF content is indexed, so results turn up files Obsidian's own search cannot find.
• Semantic search: press Tab to run the same query by meaning. Requires an embedding model, which can run locally through Ollama.
• A search that finds nothing can be handed to the agent as a question, or turned into a new note without leaving the box.

🕸️ Smart Graph

• Notes are grouped into topics automatically with Leiden community detection, drawn as tinted regions with name pills.
• Granularity moves between a few broad topics and many narrow ones; the levels nest, so a topic at one level is always part of one topic at the level below.
• Immerse: select a topic and the graph rebuilds around just those notes, revealing the finer topics inside it.
• Bridge notes and orphans are highlighted. With an embedding model, the graph also draws inferred edges between notes that are about the same thing but never link to each other.
• Rendering goes through PixiJS, with clustering, projection and the neighbour scan in web workers.

🧠 Agents

• Tool-using agents that search and read your vault before answering, and cite what they used as wikilinks.
• Every note edit is staged for review as an inline diff. Nothing is written to disk until you accept it, per hunk or all at once.
• Skills: capability is defined by folders holding a SKILL.md, in the open Agent Skills format. Four core skills ship with the plugin; agents can also write their own.
• Memory: plain notes under Agents/Memories/, readable and editable like anything else in the vault.
• Chats are notes: conversations are .chat files, indexed and searchable, with branching history, and embeddable in a note with ![[Conversation.chat]].
• Integrations with Dataview, Tasks, TaskNotes, Canvas and Bases, plus auto-discovery of any plugin exposing an api.
• MCP servers per agent, over HTTP and stdio.
• Multiple agents on different models running in parallel, subagents, and image and PDF attachments.

🔒 Privacy

• Vault-wide private by default or public by default, refined per note or per folder.
• Local providers (Ollama, oMLX) are trusted; cloud providers are untrusted and blocked from reading, embedding, or being sent a private note. The block is enforced in the plugin.
• Search and the graph never send anything anywhere. No telemetry.

🔌 Providers

Ollama, oMLX, OpenAI, Anthropic, OpenRouter, and any OpenAI-compatible endpoint.

📱 Platform

Desktop and mobile. Requires Obsidian 1.11.4 or later.

⬆️ Upgrading from 1.x

The plugin migrates its data on first run: agent context is consolidated under the Agents/ folder, and the bundled core skills are seeded into your vault. Chats from 1.x are converted to the .chat format on load.

Because the retrieval stack changed, the vector index is rebuilt on first launch. It runs in the background with progress in the status bar and is incremental after that.

If you sync your vault, exclude .obsidian/plugins/smart-second-brain/ from sync to avoid version-history bloat from the index.

💬 Feedback

This is a big release, and the surest way it gets better is hearing how it goes in your vault. Bug reports and feature requests belong in issues, questions in Q&A, and if you build a skill or an agent you like, we'd genuinely love to see it in Show and tell.

We started this as a university project. That finished a long time ago — we keep building it because we use it every day. Thanks for trying it.