LlamaDB

High-performance vector database optimized for AI workloads with MLX acceleration for Apple Silicon.

Overview

LlamaDB is a next-generation vector database designed for AI applications with a focus on performance, ease of use, and platform-specific optimizations. It provides:

• 🚀 High-performance vector search - Fast similarity search for embeddings
• 🍎 MLX acceleration - Up to 10x speedup on Apple Silicon devices
• 🦀 Rust extensions - Critical paths implemented in Rust for maximum performance
• 🐍 Python-first API - Simple, intuitive API with Python at its core
• 🔌 REST API - Easy integration with any language or framework

Quick Start

Installation

# Clone the repository
git clone https://github.com/yourusername/llamadb.git
cd llamadb

# Set up development environment
./setup_dev_environment.sh

# Activate the virtual environment
source .venv/bin/activate

Run the Quickstart Demo

python quickstart.py

This will demonstrate basic vector operations, similarity search, and performance benchmarks.

Start the API Server

# Start the server
python api_launcher.py --start

# Check status
python api_launcher.py --status

# Stop the server
python api_launcher.py --stop

Core Features

Vector Index

The VectorIndex class provides efficient similarity search:

from llamadb.core import VectorIndex

# Create a new index with 128-dimensional vectors
index = VectorIndex(dimension=128)

# Add vectors with metadata
index.add_item(embedding, {"id": 1, "text": "Document content", "category": "Technology"})

# Search for similar vectors
results = index.search(query_vector, k=10)

Accelerated Operations

LlamaDB provides optimized vector operations:

from llamadb.core import cosine_similarity, l2_distance, dot_product

# Calculate similarity between vectors
similarity = cosine_similarity(vector_a, vector_b)
distance = l2_distance(vector_a, vector_b)
dot = dot_product(vector_a, vector_b)

MLX Acceleration

On Apple Silicon devices, LlamaDB automatically uses MLX for acceleration:

from llamadb.core import is_mlx_available, is_apple_silicon

if is_apple_silicon():
    print("Running on Apple Silicon")
    
if is_mlx_available():
    print("MLX acceleration is available")

Development

For development instructions, see DEVELOPER.md.

Utility Scripts

LlamaDB includes several utility scripts:

# Setup development environment
./setup_dev_environment.sh

# Run comprehensive tests
python test_llamadb.py --benchmark --plot

# Manage API server
python api_launcher.py --start

Command Line Utilities

The commands.sh script provides common operations:

# Show help information
./commands.sh help

# Set up development environment
./commands.sh setup

# Run tests
./commands.sh test

# Run benchmarks
./commands.sh benchmark

# Run quickstart demo
./commands.sh quickstart

API Server

LlamaDB includes a REST API server for language-agnostic access:

Start the server

python api_launcher.py --start

API Endpoints

• GET /health - Health check
• GET /system - System information
• GET /index - Index information
• POST /search - Search for similar vectors

Example search request:

{
  "vector": [0.1, 0.2, 0.3, ...],
  "k": 10
}

Performance

LlamaDB is designed for performance:

Operation	NumPy	MLX (Apple Silicon)	Speedup
Cosine Similarity	5µs	0.5µs	10x
Matrix Multiply (1000x1000)	50ms	5ms	10x
Search (10k vectors)	10ms	1ms	10x

License

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

Acknowledgements

• MLX - For the Apple Silicon acceleration
• NumPy - For fundamental array operations
• FastAPI - For the REST API server