codeprysm-search

Semantic code search with vector embeddings and Qdrant.

Part of the CodePrism project.

Features

• Semantic Search: Natural language queries using vector embeddings
• Hybrid Search: Combines semantic and keyword matching with score fusion
• GPU Acceleration: Metal (macOS) and CUDA (Linux/Windows) support
• Qdrant Integration: Scalable vector database for production use
• Code-Optimized Embeddings: Uses Jina embeddings tuned for code

Installation

[dependencies]
codeprysm-search = "0.1"

# With GPU acceleration
codeprysm-search = { version = "0.1", features = ["metal"] }  # macOS
codeprysm-search = { version = "0.1", features = ["cuda"] }   # Linux/Windows

Usage

use codeprysm_search::{QdrantClient, SearchQuery};

// Connect to Qdrant
let client = QdrantClient::new("http://localhost:6334").await?;

// Search for code
let results = client.search(SearchQuery {
    query: "authentication handler".to_string(),
    limit: 10,
    ..Default::default()
}).await?;

for result in results {
    println!("{}: {} (score: {:.2})", result.file_path, result.name, result.score);
}

GPU Acceleration

GPU provides 7-9x faster inference for embedding generation:

License

MIT License - see LICENSE