Skip to content

Searching Your Vault

Atlas includes a powerful built-in search engine that runs entirely on your computer. No data leaves your device, there are no usage limits, and it works whether or not you’re subscribed. Search is a core free feature.

Atlas uses hybrid search — a combination of two different search techniques applied simultaneously:

Semantic search understands meaning. It converts your notes and your query into mathematical vectors (using a local AI embedding model) and finds notes that are conceptually similar — even if they don’t share the same words.

“What did I write about feeling burned out?” will surface journal entries about exhaustion, overwork, and stress — even if you never used those exact words.

Keyword search (BM25) finds exact and near-exact matches. It’s fast, precise, and excellent at finding specific names, technical terms, identifiers, and anything where exact wording matters.

“meeting with Jordan on 2026-01-15” will reliably surface that specific meeting note.

The results from both methods are combined using Reciprocal Rank Fusion (RRF) — an algorithm that merges two ranked lists into one, rewarding notes that score well in both. A note that ranks highly in both semantic similarity and keyword relevance will surface near the top.

You can tune how much weight each method gets in Settings > Search & RAG using the vector weight and text weight sliders.

From the Files tab:

Use the search bar at the top of the Files tab. Results appear instantly as you type, showing matching files with a brief excerpt.

From chat:

Ask Atlas naturally:

“Find my notes about the Q1 product roadmap.”

“What have I written about sleep or recovery?”

“Search for meeting notes from last week.”

Atlas will run the search and present the relevant results, often pulling in context from the most relevant notes automatically.

Before Atlas can search your vault, it needs to index it. Indexing happens automatically:

  1. First run: When you set a vault path, Atlas starts indexing in the background. Larger vaults may take a few minutes.
  2. Delta indexing: After the initial index, Atlas only re-indexes files that have changed since the last time it looked — this makes re-indexing fast even for large vaults.
  3. File watcher integration: When a file is modified (in Atlas or in an external editor), Atlas re-indexes that file automatically.

You can check how many files are indexed and when the last index was built:

  • In the Files tab, look for the index status indicator
  • Ask Atlas: “What’s my index status?” or “How many files are indexed?”

If search results seem stale or incomplete, you can manually trigger a re-index:

“Reindex my vault.”

Open Settings > Search & RAG to tune search behavior:

SettingWhat it doesDefault
Chunk sizeHow many characters each indexed chunk contains500
Chunk overlapHow much adjacent chunks overlap (helps context continuity)50
Vector weightHow much semantic similarity influences ranking0.7
Text weightHow much keyword matching influences ranking0.3
Embedding modelThe local ONNX model used for semantic embeddingsBGE

If your index gets corrupted, becomes very large, or you want to start fresh:

“Clear my search index.”

Or use the Clear Index button in Settings > Search & RAG. This deletes the stored index and triggers a full re-index from scratch.

  • Short, specific queries work well — “Jordan meeting March” is often better than a long sentence
  • Semantic search shines for fuzzy recall — you don’t need to remember exact titles or phrasing
  • Keyword search wins for exact matches — names, dates, IDs, technical terms
  • Hybrid finds both — the combined ranking means you rarely need to think about which mode to use; just ask