Semantic Search

Standard keyword search is fast and precise, but it only finds exact matches. If you search for “vacation planning,” it won’t surface a note titled “Trip Itinerary for Greece” or one that mentions “holiday preparation.” Semantic search closes that gap by understanding what your words mean, not just what they say.

What Semantic Search Does

Semantic search finds notes based on meaning and conceptual similarity. It understands that:

• “vacation planning” is related to “trip itinerary” and “holiday preparation”
• “debugging performance issues” connects to “slow API response times”
• “hiring process” is relevant to “interview questions” and “onboarding checklist”

This means your searches surface notes you’d otherwise miss — the ones that are relevant but don’t happen to use the exact words you typed.

How to Enable It

1. Open Settings (Cmd+, / Ctrl+,).
2. Navigate to AI & Intelligence.
3. Enable Semantic Search (ai.semanticSearch).

On first enable, Ariv downloads a small AI model (~22MB) that runs locally on your machine. This is a one-time download.

Note

Semantic search runs entirely on your machine using a local embedding model. No API key is required, no data leaves your computer, and it works offline after the initial model download. It doesn’t count against any usage limits.

How It Works

Behind the scenes, semantic search uses a technique called vector embeddings:

1. Indexing: When you create or modify a note, Ariv generates a vector embedding — a numerical representation of the note’s meaning. This embedding is stored locally alongside your vault’s search index.
2. Querying: When you search, your search query is also converted into a vector embedding.
3. Matching: Ariv uses cosine similarity to compare your query’s embedding against every note’s embedding, finding the ones that are closest in meaning.
4. Results: Notes are ranked by semantic similarity and blended with keyword search results for the best of both worlds.

The result is a search experience that combines the precision of keyword matching with the intelligence of meaning-based retrieval.

Where Semantic Search Helps

Semantic search enhances two key areas in Ariv:

Manual Search

When you use the search dialog (Cmd+K / Ctrl+K), semantic search results are blended with traditional full-text results. You get keyword-exact matches alongside conceptually related notes, all in one result set.

Ask Brain

Semantic search significantly improves Ask Brain by helping it find the most relevant context for your questions. When you ask a question, Ariv uses semantic matching to retrieve notes that are conceptually related to your query — even if they don’t share exact keywords. This leads to more complete, accurate answers.

Tip

Semantic search is most impactful in larger vaults (100+ notes) where keyword search alone starts to miss relevant connections. If your vault has a few dozen notes, keyword search is usually sufficient. But as your vault grows, semantic search becomes increasingly valuable.

Performance

Semantic search is designed to stay out of your way:

• Background indexing: Embeddings are generated in the background as you create and edit notes. You won’t notice any slowdown while writing.
• Incremental updates: Only new or modified notes need re-indexing. Ariv doesn’t re-process your entire vault every time.
• Small model: The local embedding model is roughly 22MB — small enough to download quickly and light enough to run without impacting system performance.
• No GPU required: The model runs on your CPU. No special hardware is needed.

First-Time Setup

When you enable semantic search for the first time:

1. The embedding model (~22MB) downloads automatically.
2. Ariv begins indexing your existing notes in the background.
3. Depending on vault size, initial indexing may take a few minutes for large vaults (1,000+ notes).
4. Once indexing is complete, semantic search is fully active.

You can continue using Ariv normally while indexing runs. Keyword search remains fully available during this process.

Caution

If you have a very large vault and limited system resources, initial indexing may temporarily increase CPU usage. This settles down once the backlog is processed and only incremental updates are needed.

Privacy

Since semantic search uses a local model, your note content is never sent to any external service. Embeddings are computed on your machine and stored alongside your vault’s local search index. This is true regardless of which AI provider you have configured for other features — semantic search always stays local.

---

Related: AI Setup — configure AI features in Ariv | Auto-Tagging — another AI-enhanced feature | Search — full overview of Ariv’s search capabilities