vowdb 0.1.6

A high-performance vector database for fast text similarity search

VowDB 🔥

Blazing-fast vector database for similarity search, like Pinecone or Milvus. Powered by Faiss & any embedding model.

Install 💻

Install the latest version of VowDB using pip:

pip install vowdb

Requires Python 3.8+, faiss-cpu, numpy, psutil, and an embedding model (e.g., sentence-transformers, langchain_ollama).

Setup 🚀

Initialize with any embedding model and specify vector dimension. For Ollama, ensure the server is running and the model is pulled.

Ollama Setup

Install Ollama: Download from https://ollama.com/ and follow installation instructions.

Start the Ollama server:

ollama serve

Pull the model:

ollama pull nomic-embed-text

Code Example

from sentence_transformers import SentenceTransformer from langchain_ollama import OllamaEmbeddings from vowdb import VowDB

Example with SentenceTransformers (dimension: 384)

model = SentenceTransformer("all-MiniLM-L6-v2", cache_folder="./model_cache") db = VowDB(embedding_model=model, vector_dim=384, file_path="vectors.faiss")

Example with OllamaEmbeddings (dimension: 768 for nomic-embed-text)

ollama_model = OllamaEmbeddings(model="nomic-embed-text") db_ollama = VowDB(embedding_model=ollama_model, vector_dim=768, file_path="ollama_vectors.faiss")

Ready in ~1.2s.

Add Data 📝

Insert vectors with metadata:

One

result = db.insert("Hello World", metadata={"greeting": "yes"}) # ~0.57s

Returns: {"status": "inserted", "id": 0, "vector": "[0.1,0.2,...]", "metadata": {"greeting": "yes", "id": "uuid", ...}}

Many

texts = ["Hii", "I Am Rushikesh!", "News article"] metadatas = [{"greeting": "yes"}, {"category": "introduction"}, {"greeting": "no"}] results = db.insert_batch(texts, metadatas)

Search 🔍

Query with vectors and SQL-like filters (use && for AND, || for OR). Returns vectors and metadata.

1. AND Power

db.find("hello", top_k=3, filter_query="category=news && score>0.9 && greeting=yes")

Result: id=0, distance=0.45, vector=[0.1,0.2,...], metadata={category=news, score=0.95, greeting=yes}

2. Vector Search

db.find("hello", top_k=3)

Results: id=0, distance=0.45, vector=[0.1,0.2,...], metadata={category=news, score=0.95, greeting=yes} id=1, distance=0.48, vector=[0.3,0.4,...], metadata={category=greeting, score=0.6, greeting=yes}

3. Exact Match

db.find("Hello World", top_k=3, filter_query="text=Hello World")

Result: id=0, distance=0.45, vector=[0.1,0.2,...], metadata={category=news, score=0.95, greeting=yes}

4. Prefix Hunt

db.find("Hel", top_k=3, filter_query="text=Hel*")

Results: id=0, distance=0.45, vector=[0.1,0.2,...], metadata={category=news, score=0.95, greeting=yes} id=1, distance=0.48, vector=[0.3,0.4,...], metadata={category=greeting, score=0.6, greeting=yes}

5. Category Snap

db.find("news", top_k=3, filter_query="category=news")

Results: id=0, distance=0.45, vector=[0.1,0.2,...], metadata={category=news, score=0.95, greeting=yes} id=2, distance=0.50, vector=[0.5,0.6,...], metadata={category=news, score=0.7, greeting=no}

6. OR Flex

db.find("news", top_k=3, filter_query="category=news||greeting")

Results: id=0, distance=0.45, vector=[0.1,0.2,...], metadata={category=news, score=0.95, greeting=yes} id=1, distance=0.48, vector=[0.3,0.4,...], metadata={category=greeting, score=0.6, greeting=yes}

7. Multi-Filter

db.find("news", top_k=3, filter_query="category=news && score>0.7 && greeting=yes")

Result: id=0, distance=0.45, vector=[0.1,0.2,...], metadata={category=news, score=0.95, greeting=yes}

8. Score Range

db.find("news", top_k=3, filter_query="score=0.6-0.8")

Results: id=1, distance=0.48, vector=[0.3,0.4,...], metadata={category=greeting, score=0.6, greeting=yes} id=2, distance=0.50, vector=[0.5,0.6,...], metadata={category=news, score=0.7, greeting=no}

9. Skip Some

db.find("news", top_k=3, filter_query="text!=Hii && category!=introduction")

Results: id=0, distance=0.45, vector=[0.1,0.2,...], metadata={category=news, score=0.95, greeting=yes} id=2, distance=0.50, vector=[0.5,0.6,...], metadata={category=news, score=0.7, greeting=no}

10. Name Drop

db.find("Rushikesh", top_k=3)

Result: id=1, distance=0.20, vector=[0.7,0.8,...], metadata={category=introduction, score=0.9}

11. Text or Score

db.find("Hel", top_k=3, filter_query="text=Hel*||score>=0.9")

Results: id=0, distance=0.45, vector=[0.1,0.2,...], metadata={category=news, score=0.95, greeting=yes} id=1, distance=0.20, vector=[0.7,0.8,...], metadata={category=introduction, score=0.9}

Save/Load 💾

Keep or get data:

db.save() db.load()

Info 🌟

Version: 0.1.5 Author: Rushikesh Sunil Kotkar License: MIT GitHub: https://github.com/rushikeshkotkar04/vowdb PyPI: Fast vector similarity with slick = queries.

Contribute 🤝

Got ideas? Issues? PRs? Hit up: https://github.com/rushikeshkotkar04/vowdb