README.md

Mistral VectorDB

A high-performance vector database optimized for Mistral AI embeddings, featuring efficient similarity search, storage, and retrieval capabilities.

Features

• Optimized for Mistral AI: Built specifically for Mistral's embedding model
• Efficient Vector Search: Uses FAISS with HNSW and IVF for fast similarity search
• Advanced Storage: Compressed storage with LSM-tree inspired design
• Rich Querying: Metadata filtering and customizable search parameters
• Batch Processing: Efficient handling of bulk operations
• Caching System: Smart caching for frequently accessed embeddings
• Easy Integration: Simple API for seamless integration

Installation

pip install mistral-vectordb

Quick Start

from mistral_vectordb import VectorDatabase, MistralEmbeddings

# Initialize with your Mistral API key
embeddings = MistralEmbeddings(api_key="your-api-key")
db = VectorDatabase("db_path", dimension=embeddings.dimension)

# Add documents
text = "Sample document"
embedding = embeddings.embed(text)
doc_id = db.add_document(
    content=text,
    embedding=embedding[0],
    metadata={"category": "tech"}
)

# Search
query = "similar document"
query_embedding = embeddings.embed(query)
results = db.search(
    query_embedding=query_embedding[0],
    k=10,
    threshold=0.7,
    metadata_filters={"category": "tech"}
)

Advanced Usage

Batch Processing

# Embed multiple documents
texts = ["Document 1", "Document 2", "Document 3"]
embeddings_array = embeddings.bulk_embed(
    texts,
    batch_size=32,
    show_progress=True
)

# Add to database
for text, embedding in zip(texts, embeddings_array):
    db.add_document(
        content=text,
        embedding=embedding,
        metadata={"batch": "example"}
    )

Custom Search Parameters

results = db.search(
    query_embedding=query_embedding[0],
    k=5,                    # Number of results
    threshold=0.8,          # Minimum similarity score
    metadata_filters={      # Filter by metadata
        "category": "tech",
        "language": "en"
    }
)

API Reference

MistralEmbeddings

embeddings = MistralEmbeddings(
    api_key="your-api-key",
    model="mistral-embed",    # Embedding model to use
    cache_dir="path/to/cache", # Optional cache directory
    cache_duration=24         # Cache duration in hours
)

# Generate embeddings
embedding = embeddings.embed("text")
embeddings_array = embeddings.bulk_embed(["text1", "text2"])

VectorDatabase

db = VectorDatabase(
    path="db_path",           # Database storage path
    dimension=1024            # Embedding dimension
)

# Add document
doc_id = db.add_document(
    content="text",           # Original text
    embedding=vector,         # NumPy array
    metadata={"key": "value"} # Optional metadata
)

# Search
results = db.search(
    query_embedding=vector,   # Query vector
    k=10,                    # Number of results
    threshold=0.7,           # Similarity threshold
    metadata_filters={}      # Optional filters
)

Contributing

Contributions are welcome! Please read our Contributing Guidelines for details on how to submit pull requests, report issues, and contribute to the project.