vecstream 0.1.0

A lightweight, efficient vector database with similarity search capabilities

VecStream

A lightweight, efficient vector database with similarity search capabilities, optimized for machine learning applications.

Features

• 🚀 Fast in-memory vector storage and retrieval
• 🔍 Semantic similarity search using cosine similarity
• 📊 Built-in text embedding using Sentence Transformers
• 💻 Clean CLI interface for easy interaction
• 🛠 Python API for programmatic access

Installation

pip install vecstream

CLI Usage

VecStream provides a command-line interface for common operations:

Add a text entry

vecstream add "Your text here" text_id_1

Search for similar entries

vecstream search "Query text" --k 5 --threshold 0.5

Get vector by ID

vecstream get text_id_1

Remove vector

vecstream remove text_id_1

Show database info

vecstream info

Clear database

vecstream clear

Python API Usage

from vecstream import VectorStore, IndexManager, QueryEngine
from sentence_transformers import SentenceTransformer

# Initialize components
store = VectorStore()
index_manager = IndexManager(store)
query_engine = QueryEngine(index_manager)
model = SentenceTransformer('all-MiniLM-L6-v2')

# Add vectors
text = "Example text"
vector = model.encode(text)
store.add("doc1", vector)

# Search
query_vector = model.encode("Search query")
results = query_engine.search(query_vector, k=5)
for id, similarity in results:
    print(f"Match {id}: {similarity:.4f}")

Performance

• Fast query response times (typically < 10ms)
• Efficient memory usage
• Linear scaling with dataset size
• Support for concurrent queries

Technical Details

• Uses Sentence Transformers for text embedding
• 384-dimensional vectors by default
• Cosine similarity for vector comparison
• In-memory storage with optional persistence
• Rich CLI interface with progress indicators

Requirements

• Python 3.8+
• numpy
• scipy
• scikit-learn
• sentence-transformers
• click
• rich

License

MIT License

Author

Torin Etheridge