FastEmbed VectorStore

A high-performance, Rust-based in-memory vector store with FastEmbed integration for Python applications.

Overview

FastEmbed VectorStore is a lightweight, fast vector database that leverages the power of Rust and the FastEmbed library to provide efficient text embedding and similarity search capabilities. It's designed for applications that need quick semantic search without the overhead of external database systems.

Features

🚀 High Performance: Built in Rust with Python bindings for optimal speed
🧠 Multiple Embedding Models: Support for 30+ pre-trained embedding models including BGE, Nomic, GTE, and more
💾 In-Memory Storage: Fast in-memory vector storage with persistence capabilities
🔍 Similarity Search: Cosine similarity-based search with customizable result limits
💾 Save/Load: Persist and restore vector stores to/from JSON files
🐍 Python Integration: Seamless Python API with PyO3 bindings

Supported Embedding Models

The library supports a wide variety of embedding models:

BGE Models: BGEBaseENV15, BGELargeENV15, BGESmallENV15 (with quantized variants)
Nomic Models: NomicEmbedTextV1, NomicEmbedTextV15 (with quantized variants)
GTE Models: GTEBaseENV15, GTELargeENV15 (with quantized variants)
Multilingual Models: MultilingualE5Small, MultilingualE5Base, MultilingualE5Large
Specialized Models: ClipVitB32, JinaEmbeddingsV2BaseCode, ModernBertEmbedLarge
And many more...

Installation

Prerequisites

Python 3.8 or higher
Rust toolchain (to build from source)

Install from PyPI

pip install fastembed-vectorstore

From Source

Clone the repository:

git clone https://github.com/sauravniraula/fastembed_vectorstore.git
cd fastembed_vectorstore

Install the package:

maturin develop

Quick Start

from fastembed_vectorstore import FastembedVectorstore, FastembedEmbeddingModel

# Initialize with a model
model = FastembedEmbeddingModel.BGESmallENV15
vectorstore = FastembedVectorstore(model)

# Add documents
documents = [
    "The quick brown fox jumps over the lazy dog",
    "A quick brown dog jumps over the lazy fox",
    "The lazy fox sleeps while the quick brown dog watches",
    "Python is a programming language",
    "Rust is a systems programming language"
]

# Embed and store documents
success = vectorstore.embed_documents(documents)
print(f"Documents embedded: {success}")

# Search for similar documents
query = "What is Python?"
results = vectorstore.search(query, n=3)

for doc, similarity in results:
    print(f"Document: {doc}")
    print(f"Similarity: {similarity:.4f}")
    print("---")

# Save the vector store
vectorstore.save("my_vectorstore.json")

# Load the vector store later
loaded_vectorstore = FastembedVectorstore.load(model, "my_vectorstore.json")

API Reference

FastembedEmbeddingModel

Enum containing all supported embedding models. Choose based on your use case:

Small models: Faster, lower memory usage (e.g., BGESmallENV15)
Base models: Balanced performance (e.g., BGEBaseENV15)
Large models: Higher quality embeddings (e.g., BGELargeENV15)
Quantized models: Reduced memory usage (e.g., BGESmallENV15Q)