sqlite-vector-search

README

SQLite Vector Search with Ollama Embeddings

This experiment demonstrates SQLite vector search capabilities using custom functions and real embeddings from Ollama.

Features

• Custom SQLite functions for cosine similarity calculation
• Integration with Ollama for generating embeddings using all-minilm:latest model
• CLI interface for adding documents and searching
• Sample data insertion and demonstration

Prerequisites

• Ollama running locally on 127.0.0.1:11434
• all-minilm:latest model available in Ollama

Usage

Basic Demo

Run the default demo which inserts sample documents and performs a search:

go run ./cmd/experiments/sqlite-vector-search

Add Documents

Add a new document to the database:

go run ./cmd/experiments/sqlite-vector-search add "Your document text here"

Search Documents

Search for similar documents:

go run ./cmd/experiments/sqlite-vector-search search "your search query" --limit 5

Options

• --log-level: Set log level (debug, info, warn, error)
• --ollama-url: Ollama server URL (default: http://127.0.0.1:11434)
• --ollama-model: Ollama model to use (default: all-minilm:latest)
• --db-path: SQLite database path (default: vector_search.db)

How It Works

1. Custom SQLite Functions: Registers a cosine_similarity function in SQLite
2. Embedding Generation: Uses Ollama API to generate embeddings for text
3. Vector Storage: Stores embeddings as JSON in SQLite
4. Similarity Search: Calculates cosine similarity between query and stored embeddings

Example Output

Search results for: artificial intelligence and neural networks
============================================================
1. [0.8234] Machine learning is a subset of artificial intelligence
2. [0.7891] Deep learning uses neural networks with multiple layers
3. [0.6456] Natural language processing involves understanding text
4. [0.5234] Embeddings capture semantic meaning of text
5. [0.3456] Vector databases are useful for similarity search

Technical Details

• Uses github.com/mattn/go-sqlite3 for SQLite integration
• Implements cosine similarity calculation in Go
• JSON encoding for vector storage in SQLite
• Zerolog for structured logging
• Cobra for CLI interface