First run
Smart Second Brain is built in three layers, and each one is useful on its own. Nothing below requires an account, an API key, or an internet connection until you reach the agent.
Works immediately — no provider
Section titled “Works immediately — no provider”Smarter search. Full-text, fuzzy, and tag/frontmatter-aware. Already more capable than Obsidian’s built-in search, with no setup.
Smart Graph. An interactive graph view with clustering, layout, and filtering. Out of the box it renders your wikilinks.
Open either from the command palette:
Smart Second Brain: Search notesSmart Second Brain: Open graph
Add an embedding model — semantic layer
Section titled “Add an embedding model — semantic layer”An embedding model unlocks concept-level retrieval: search matches meaning rather than exact words, and the graph gains semantic similarity edges between notes that aren’t explicitly linked.
This can run fully locally through Ollama — a strong
starting point is mxbai-embed-large. No note content leaves your machine.
Configure it under Settings → Smart Second Brain → Embedding model.
Add an AI provider — the agent
Section titled “Add an AI provider — the agent”A chat model enables the full agent: it searches your vault, reads what’s relevant, and answers with links back to the notes it used. See Connecting a provider.
Indexing
Section titled “Indexing”The first index build runs in the background and its progress appears in the status bar. Large vaults take a few minutes. Indexing is incremental after that — only changed notes are reprocessed.