Search
Search is the layer that works the moment you install the plugin. No provider, no API key, no index build to wait through — open the command palette and run Smart Second Brain: Search notes.
Two retrieval paths
Section titled “Two retrieval paths”Search runs a lexical path and, when an embedding model is configured, a semantic path, then fuses their results into one ranking.
Lexical matches the words you typed. It is BM25-style scoring over a MiniSearch index, with fuzzy matching for typos and prefix matching so partial words still hit. It needs no model and no network.
Semantic matches meaning. Your query is embedded and compared against the embedded chunks of your notes, so a search for “why my mornings feel scattered” can surface a note that only ever says “context switching”. This path requires an embedding model — see First run.
Results from both paths are normalized, fused, and re-ranked before you see them. The search algorithm reference documents the pipeline in full, including the constants.
Choosing a strategy
Section titled “Choosing a strategy”Search exposes three strategies:
| Strategy | What it matches | Needs an embedding model |
|---|---|---|
lexical |
Exact and fuzzy words | No |
semantic |
Meaning and concepts | Yes |
hybrid |
Both, fused | Yes |
Reach for lexical when you know the vocabulary — a project code name, a
person, an error string. Reach for semantic when you know the idea but not
the words your past self used. hybrid is the sensible default once embeddings
are configured, and is what the agent escalates to when a query mixes an exact
term with a fuzzy concept.
Filters
Section titled “Filters”Searches can be narrowed before ranking rather than after:
- Path — restrict to a folder or a single note.
- Tags — restrict to notes carrying given tags.
- Properties — restrict on frontmatter keys and values.
Filters apply at chunk level on the semantic path, so a filtered semantic search doesn’t quietly waste its result budget on chunks that were going to be discarded anyway.
Recency
Section titled “Recency”Notes you have touched recently get a ranking boost. This is deliberate and mild: it breaks ties in favour of what you’re actually working on without letting a freshly-saved but irrelevant note outrank a strong match.
What search excludes
Section titled “What search excludes”The agent’s own folder — Agents/ by default, holding memories, skills, and
system prompts — is excluded from indexing, search, and the graph. It is plugin
machinery that happens to be stored as notes, and it would otherwise dominate
results about the plugin itself.
Binary files such as images and PDFs are indexed by title only. Their content is not embedded.
Chat files
Section titled “Chat files”Conversations are stored as .chat files in your vault, and they are indexed —
so you can find a past conversation by searching for what was said in it.
Because conversation history is a tree of checkpoints where each checkpoint
re-contains every prior message, chat files are deduped by message id before
indexing. Without that, a single conversation would be indexed many times over.
Chats are also transcludable — ![[My conversation.chat]] renders the
transcript inside another note. See
Embedding a chat in a note.
The search_notes tool
Section titled “The search_notes tool”The agent reaches the same pipeline through its search_notes tool, with the
same strategies and filters. When you ask the agent a question about your
notes, this is what runs. See Agents.