CodePrysm

A powerful tool for analyzing code repositories and generating relationship graphs using Tree-sitter abstract syntax trees. CodePrysm transforms source code into a searchable knowledge graph, enabling semantic code search, dependency analysis, and intelligent code navigation.

Overview

CodePrysm builds a comprehensive graph representation of your codebase where:

• Nodes represent code entities using three semantic types: Container (classes, interfaces, structs), Callable (functions, methods), and Data (fields, properties, constants)
• Edges represent three types of relationships: CONTAINS (hierarchy), USES (dependencies), and DEFINES (definitions)
• Embeddings enable semantic search using natural language with full metadata support

Key Features

• Semantic Code Search - Find code using natural language queries with kind/subtype filtering
• AST-Based Analysis - Precise parsing using Tree-sitter with declarative SCM tags
• Rich Dependency Graphs - Three relationship types (CONTAINS, USES, DEFINES) for comprehensive analysis
• Fine-Grained Entities - Distinguish structs from interfaces, async from sync, fields from properties
• Scalable Architecture - Handles codebases with 100K+ files
• MCP Integration - AI-powered code exploration via Model Context Protocol
• Multi-Language - Python, JavaScript/TypeScript, C/C++, C#, Go, Rust
• GPU Acceleration - Metal (macOS) and CUDA (Linux/Windows) support

Installation

From crates.io (Recommended)

cargo install codeprysm-cli

From Source

git clone https://github.com/codeprysm/codeprysm.git
cd codeprysm
cargo build --release

GPU Acceleration (Optional)

For faster embedding generation, install with GPU support:

# macOS (Apple Silicon)
cargo install codeprysm-cli --features metal

# Linux (NVIDIA GPU)
cargo install codeprysm-cli --features cuda

Prerequisites

• Docker (for Qdrant vector database)
• Rust 1.85+ (only if building from source)

Quick Start

1. Start Qdrant (required for semantic search):

   docker run -d --name qdrant \
     -p 6333:6333 -p 6334:6334 \
     -v qdrant_storage:/qdrant/storage \
     qdrant/qdrant:latest
   

2. Initialize your codebase:

   cd /path/to/your/repo
   codeprysm init
   

3. Start the MCP server (optional, for AI assistants):

   codeprysm mcp
   

How It Works

graph LR
    A[Source Code] --> B[Code Graph]
    B --> C[Semantic Index]
    C --> D[MCP Server]
    D --> E[AI Assistants]

The system operates in three main phases:

1. Code Graph Generation - Parse source files into a graph structure using Tree-sitter AST
2. Indexing for Search - Create embeddings for semantic search using Qdrant
3. MCP Server Integration - Expose capabilities to AI assistants via MCP protocol

Documentation

• Getting Started - Setup guide
• Docker Setup - Running with Docker
• SCM Tag Convention - Query file syntax
• SCM Overlays - Adding scope metadata

CLI Commands

# Generate code graph
codeprysm init --root /path/to/repo

# Start MCP server
codeprysm mcp --root /path/to/repo --qdrant-url http://localhost:6334

# Search codebase
codeprysm search "function that handles authentication"

# Show statistics
codeprysm stats --codeprysm-dir .codeprysm

# Incremental update
codeprysm update --root /path/to/repo

Supported Languages

Performance & Scalability

Development

For development, install just command runner:

# Build
just rust-build

# Test
just rust-test

# Lint
just rust-lint

# Format
just rust-fmt

See CONTRIBUTING.md for detailed development guidelines.

Project Structure

codeprysm/
├── crates/
│   ├── codeprysm-core/     # Graph generation, tree-sitter parsing
│   ├── codeprysm-search/   # Vector search, embeddings
│   ├── codeprysm-mcp/      # MCP server
│   ├── codeprysm-cli/      # Command-line interface
│   ├── codeprysm-config/   # Configuration management
│   └── codeprysm-backend/  # Backend abstraction
├── tests/fixtures/         # Test repositories
├── docs/                   # Documentation
└── docker/                 # Docker configuration

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for guidelines.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Tree-sitter for powerful parsing capabilities
• Qdrant for vector search
• Candle for ML inference