AI Integration

Semantic Search & Embeddings

Understand how vector embeddings power Snipset's OmniSearch.

This page covers the technical underpinning of OmniSearch: vector embeddings. If you haven't set up OmniSearch yet, start with the OmniSearch feature page . This guide goes deeper into how the embedding system works and how to manage your database.

What Are Vector Embeddings?

Computers don't understand the "meaning" of words organically. To solve this, AI uses vector embeddings.

An embedding is a mathematical representation of text meaning as a list of numbers (a "vector").
Two texts with similar meanings will have similar mathematical vectors, even if they use completely different words.
For example: "I need to cancel my subscription" and "How do I unsubscribe?" will have very similar embeddings, allowing the search engine to link them.
Snipset uses a local embedding model (for example hf.co/lmstudio-community/embeddinggemma-300m-qat-GGUF) running via Ollama to generate these vectors.

The Embedding Database

Because calculating embeddings requires AI processing, Snipset doesn't do it on the fly while you search. Instead, it pre-calculates them and stores them.

Embeddings are stored in a dedicated vector table inside your local SQLite database ( snipset.db).
Each snippet gets exactly one embedding entry. This entry is a combination of the snippet's Name, Keyword, and Content.
The database footprint is very small—it grows at approximately 1.5KB per snippet.

Sync Controls

In the Settings → AI & Models → AI Configuration panel, you have precise control over the embedding generation process:

Action	What it does
Start Embedding	Begins embedding all snippets that don't yet have a vector entry. It is completely safe to run this at any time.
Pause	Temporarily stops the sync. The current progress is saved, allowing you to resume whenever you like.
Resume	Continues the process from exactly where it was paused. It only processes the remaining un-embedded snippets.
Force All	Deletes ALL existing embeddings in the database and re-generates them from scratch.

Keeping Embeddings Up to Date

Automatic Updates: When you create a new snippet or edit an existing one, Snipset automatically updates its embedding in the background. You do not need to manually run a sync every time you add a snippet!

You only need to use the Force All button in specific scenarios:

If you performed a massive CSV/JSON import of hundreds of snippets.
If you changed the Embedding Model in the AI Configuration settings.
If semantic search results seem inaccurate or out of date.

Performance Considerations

Generating embeddings requires your CPU or GPU to process data via Ollama.

For large libraries (e.g., 500+ snippets), a Force Sync may take several minutes.
The Pause/Resume system ensures you're never locked out. You can interrupt the sync at any time if your computer becomes sluggish.
No typing impact: Snipset's core text-expansion engine runs on a separate, high-priority thread. Even while a massive sync is running in the background, your snippets will continue to trigger instantly without lag.

← PreviousAI Snippet Generation Next →Ollama A-Z Tutorial