vittoriadb 0.4.0

Simple embedded vector database for local AI development with automatic embeddings

VittoriaDB Python SDK

VittoriaDB Python SDK is a client library for VittoriaDB, a high-performance embedded vector database built in Go. This SDK provides a clean, Pythonic interface to interact with VittoriaDB server instances, with automatic binary management and server lifecycle control.

🏗️ Architecture

VittoriaDB consists of two components:

• 🚀 VittoriaDB Server (Go binary): High-performance vector database engine
• 🐍 Python SDK (this package): Client library with automatic server management

The Python SDK can either:

• Auto-manage the Go server binary (downloads, starts, stops automatically)
• Connect to an existing VittoriaDB server instance

🚀 Key Features

• 🎯 Zero Configuration: Works immediately after installation with sensible defaults
• 🤖 Automatic Embeddings: Server-side text vectorization with multiple model support
• 📄 Document Processing: Built-in support for PDF, DOCX, TXT, MD, and HTML files
• 🔧 Auto Binary Management: Automatically downloads and manages VittoriaDB binaries
• ⚡ High Performance: HNSW indexing provides sub-millisecond search times
• 🐍 Pythonic API: Clean, intuitive Python interface with type hints
• 🔌 Dual Mode: Works with existing servers or auto-starts local instances
• 🤖 RAG-Ready (NEW v0.4.0): Built-in content storage for Retrieval-Augmented Generation

📦 Installation

pip install vittoriadb

The package automatically downloads the appropriate VittoriaDB binary for your platform during installation.

🚀 Server Management

Automatic Server Management (Recommended)

import vittoriadb

# SDK automatically downloads, starts, and manages the VittoriaDB server
db = vittoriadb.connect()  # auto_start=True by default
# ... use the database ...
db.close()  # Automatically stops the server

Manual Server Management

# Download VittoriaDB binary manually
# From: https://github.com/antonellof/VittoriaDB/releases

# Start server manually
./vittoriadb run --port 8080 --data-dir ./data

# In Python, connect to existing server
import vittoriadb
db = vittoriadb.connect(url="http://localhost:8080", auto_start=False)

Connection Options

# Auto-start with custom configuration
db = vittoriadb.connect(
    auto_start=True,
    port=9090,
    host="localhost", 
    data_dir="./my_vectors"
)

# Connect to remote server
db = vittoriadb.connect(
    url="http://remote-server:8080",
    auto_start=False
)

🚀 Quick Start

Basic Usage

import vittoriadb

# Auto-starts VittoriaDB server and connects
db = vittoriadb.connect()

# Create a collection
collection = db.create_collection(
    name="documents",
    dimensions=384,
    metric="cosine"
)

# Insert vectors with metadata
collection.insert(
    id="doc1",
    vector=[0.1, 0.2, 0.3] * 128,  # 384 dimensions
    metadata={"title": "My Document", "category": "tech"}
)

# Search for similar vectors
results = collection.search(
    vector=[0.1, 0.2, 0.3] * 128,
    limit=5,
    include_metadata=True
)

for result in results:
    print(f"ID: {result.id}, Score: {result.score:.4f}")
    print(f"Metadata: {result.metadata}")

# Close connection
db.close()

Automatic Text Embeddings (🚀 NEW!)

import vittoriadb
from vittoriadb.configure import Configure

# Connect to VittoriaDB
db = vittoriadb.connect()

# Create collection with automatic embeddings
collection = db.create_collection(
    name="smart_docs",
    dimensions=384,
    vectorizer_config=Configure.Vectors.auto_embeddings()  # 🎯 Server-side embeddings!
)

# Insert text directly - embeddings generated automatically!
collection.insert_text(
    id="article1",
    text="Artificial intelligence is transforming how we process data.",
    metadata={"category": "AI", "source": "blog"}
)

# Batch insert multiple texts
texts = [
    {
        "id": "article2",
        "text": "Machine learning enables computers to learn from data.",
        "metadata": {"category": "ML"}
    },
    {
        "id": "article3", 
        "text": "Vector databases provide efficient similarity search.",
        "metadata": {"category": "database"}
    }
]
collection.insert_text_batch(texts)

# Search with natural language queries
results = collection.search_text(
    query="artificial intelligence and machine learning",
    limit=3
)

for result in results:
    print(f"Score: {result.score:.4f}")
    print(f"Text: {result.metadata['text'][:100]}...")

db.close()

Document Upload and Processing

import vittoriadb
from vittoriadb.configure import Configure

db = vittoriadb.connect()

# Create collection with vectorizer for automatic processing
collection = db.create_collection(
    name="knowledge_base",
    dimensions=384,
    vectorizer_config=Configure.Vectors.auto_embeddings()
)

# Upload and process documents automatically
result = collection.upload_file(
    file_path="research_paper.pdf",
    chunk_size=600,
    chunk_overlap=100,
    metadata={"source": "research", "year": "2024"}
)

print(f"Processed {result['chunks_created']} chunks")
print(f"Inserted {result['chunks_inserted']} vectors")

# Search the uploaded content
results = collection.search_text(
    query="machine learning algorithms",
    limit=5
)

db.close()

🤖 RAG (Retrieval-Augmented Generation) Support (NEW v0.4.0)

VittoriaDB now includes built-in support for RAG systems by automatically storing original text content alongside vector embeddings.

Content Storage Features

• ✅ Automatic Content Preservation: Original text stored with vectors
• ✅ No External Storage Required: Self-contained RAG solution
• ✅ Configurable Limits: Control storage size and behavior
• ✅ Fast Retrieval: Single query returns both vectors and content

RAG-Optimized Collection

import vittoriadb
from vittoriadb import ContentStorageConfig
from vittoriadb.configure import Configure

db = vittoriadb.connect()

# Create RAG-optimized collection with content storage
collection = db.create_collection(
    name="rag_documents",
    dimensions=384,
    vectorizer_config=Configure.Vectors.auto_embeddings(),
    content_storage=ContentStorageConfig(
        enabled=True,           # Store original content
        field_name="_content",  # Metadata field name
        max_size=1048576,      # 1MB limit per document
        compressed=False       # Compression (future feature)
    )
)

# Insert documents - content automatically preserved
collection.insert_text(
    id="doc1",
    text="VittoriaDB is a high-performance vector database perfect for RAG applications...",
    metadata={"title": "VittoriaDB Guide", "category": "documentation"}
)

# Search with content retrieval for RAG
results = collection.search_text(
    query="vector database RAG",
    limit=5,
    include_content=True  # Retrieve original content for LLM context
)

# Use results for RAG
for result in results:
    print(f"Score: {result.score:.3f}")
    print(f"Content: {result.content}")  # Original text for LLM context
    print(f"Has content: {result.has_content()}")

db.close()

RAG Workflow Example

# 1. Store knowledge base
documents = [
    "VittoriaDB supports automatic embeddings...",
    "RAG systems combine retrieval and generation...",
    "Vector databases enable semantic search..."
]

for i, doc in enumerate(documents):
    collection.insert_text(f"kb_{i}", doc, {"type": "knowledge"})

# 2. Query with content for LLM
query = "How do vector databases work?"
results = collection.search_text(query, include_content=True)

# 3. Build context for LLM
context = "\n".join([r.content for r in results if r.has_content()])

# 4. Send to your LLM
# response = your_llm.generate(query, context)

🎛️ Vectorizer Configuration

VittoriaDB supports multiple vectorizer backends for automatic embedding generation:

Sentence Transformers (Default)

from vittoriadb.configure import Configure

config = Configure.Vectors.sentence_transformers(
    model="all-MiniLM-L6-v2",
    dimensions=384
)

OpenAI Embeddings

config = Configure.Vectors.openai_embeddings(
    api_key="your-openai-api-key",
    model="text-embedding-ada-002",
    dimensions=1536
)

HuggingFace Models

config = Configure.Vectors.huggingface_embeddings(
    api_key="your-hf-token",  # Optional for public models
    model="sentence-transformers/all-MiniLM-L6-v2",
    dimensions=384
)

Local Ollama

config = Configure.Vectors.ollama_embeddings(
    model="nomic-embed-text",
    dimensions=768,
    base_url="http://localhost:11434"
)

📄 Document Processing

VittoriaDB supports automatic processing of various document formats:

Format	Extension	Status	Features
Plain Text	.txt	✅ Fully Supported	Direct text processing
Markdown	.md	✅ Fully Supported	Frontmatter parsing
HTML	.html	✅ Fully Supported	Tag stripping, metadata
PDF	.pdf	✅ Fully Supported	Multi-page text extraction
DOCX	.docx	✅ Fully Supported	Properties, text extraction

# Upload multiple document types
for file_path in ["doc.pdf", "guide.docx", "readme.md"]:
    result = collection.upload_file(
        file_path=file_path,
        chunk_size=500,
        metadata={"batch": "docs_2024"}
    )
    print(f"Processed {file_path}: {result['chunks_inserted']} chunks")

🔧 Advanced Configuration

Collection Configuration

# High-performance HNSW configuration
collection = db.create_collection(
    name="large_dataset",
    dimensions=1536,
    metric="cosine",
    index_type="hnsw",
    config={
        "m": 32,                # HNSW connections per node
        "ef_construction": 400,  # Construction search width
        "ef_search": 100        # Search width
    },
    vectorizer_config=Configure.Vectors.openai_embeddings(api_key="your-key")
)

Connection Options

# Connect to existing server
db = vittoriadb.connect(
    url="http://localhost:8080",
    auto_start=False
)

# Auto-start with custom configuration
db = vittoriadb.connect(
    auto_start=True,
    port=9090,
    data_dir="./my_vectors"
)

Search with Filtering

# Search with metadata filters
results = collection.search(
    vector=query_vector,
    limit=10,
    filter={"category": "technology", "year": 2024},
    include_metadata=True
)

# Text search with filters
results = collection.search_text(
    query="machine learning",
    limit=5,
    filter={"source": "research"}
)

📊 Performance and Scalability

• Insert Speed: >10,000 vectors/second with flat indexing, >5,000 with HNSW
• Search Speed: Sub-millisecond search times for 1M vectors using HNSW
• Memory Usage: <100MB for 100,000 vectors (384 dimensions)
• Scalability: Tested up to 1 million vectors, supports up to 2,048 dimensions

📋 API Reference

Collection Class

• insert(id, vector, metadata=None) - Insert single vector
• insert_batch(vectors) - Insert multiple vectors
• insert_text(id, text, metadata=None) - Insert text (auto-vectorized with content storage)
• insert_text_batch(texts) - Insert multiple texts (auto-vectorized with content storage)
• search(vector, limit=10, filter=None, include_content=False) - Vector similarity search
• search_text(query, limit=10, filter=None, include_content=False) - Text search with content retrieval
• upload_file(file_path, chunk_size=500, **kwargs) - Upload and process document
• get(id) - Get vector by ID
• delete(id) - Delete vector by ID
• count() - Get total vector count

VittoriaDB Class (Enhanced v0.4.0)

• connect(url=None, auto_start=True, **kwargs) - Connect to VittoriaDB
• create_collection(name, dimensions, metric="cosine", vectorizer_config=None, content_storage=None) - Create collection with content storage
• get_collection(name) - Get existing collection
• list_collections() - List all collections
• delete_collection(name) - Delete collection
• health() - Get server health status
• close() - Close connection

🤝 Contributing

We welcome contributions!

• Users: Report issues and request features on GitHub Issues
• Developers: See DEVELOPMENT.md for setup, building, and deployment instructions
• General: Check our Contributing Guide for project guidelines

📄 License

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

🔗 Links

• GitHub: https://github.com/antonellof/VittoriaDB
• PyPI: https://pypi.org/project/vittoriadb/
• Issues: https://github.com/antonellof/VittoriaDB/issues

🚀 What's Next?

• 🔍 Hybrid Search: Combine vector and keyword search
• 🔐 Authentication: User management and access control
• 🌐 Distributed Mode: Multi-node clustering support
• 📊 Analytics: Query performance monitoring and optimization
• 🎯 More Vectorizers: Support for additional embedding models
• 🗜️ Content Compression: Compress stored content to save space

📝 Changelog v0.4.0

🆕 New Features

• 🤖 Built-in Content Storage: Automatic preservation of original text for RAG applications
• 📋 ContentStorageConfig: Configurable content storage with size limits and field names
• 🔍 Enhanced Search: New include_content parameter for content retrieval
• 📊 Collection Info: Content storage configuration in collection metadata

🔧 Enhancements

• ⚡ Better RAG Support: No external storage required for RAG workflows
• 🎯 Improved API: Enhanced search methods with content retrieval options
• 📚 Updated Documentation: Comprehensive RAG examples and best practices

🔄 Backward Compatibility

• All existing APIs work unchanged
• Content storage is optional and configurable
• Default behavior maintains compatibility with v0.3.x

---

Happy building with VittoriaDB! 🚀