What it does

Smart2Brain embeds your Obsidian notes into a local vector store and retrieves relevant chunks when you ask questions, surfacing answers with clickable links back to source notes. It can run entirely offline through Ollama, or pipe queries to OpenAI’s models if you prefer capability over privacy. You can also disable RAG entirely and use it as a generic chat interface.

The interesting bit

The project started as a university assignment and now serves as an experimental playground for its authors. The RAG pipeline uses hierarchical tree summarization rather than simple chunking, and the whole stack—LangChain for orchestration, Orama for vector search—is open source, unlike some competing Obsidian AI plugins that require licenses for local model support.

Key highlights

• Local-first: Ollama integration keeps embeddings and inference on your machine
• Cited answers: responses include Obsidian links to the originating notes
• Model switching: swap between LLMs for different tasks (e.g., scientific vs. persuasive writing)
• Two chat UIs: “comfy” and “compact” views
• Persistent chat histories you can resume later

Caveats

• Output quality hinges heavily on how well-structured your vault is; messy notes yield messy answers
• The authors note that “development setup instructions” are still TBD, so contributing may require some spelunking
• Obsidian Sync users must manually exclude the vectorstores folder to avoid bloated version history

Verdict

Worth a look if you want private, cited Q&A over your own notes and don’t mind tuning your vault organization. Skip it if you need polished multi-language support or a mature plugin ecosystem—this is still a spare-time project with an ambitious roadmap.