pyenvector 1.3.0a1

A client application for performing encrypted vector search using homomorphic encryption.

What is enVector?

enVector is a high-performance encrypted vector search service that keeps both your vectors and similarity scores private during computation. The pyenvector Python SDK lets you talk to the enVector server from Python while keeping all sensitive data encrypted end to end.

pyenvector Example

This example walks through the typical pyenvector workflow for encrypted similarity search and vector database operations.

---

Install and Import

Install the SDK and import it as ev for brevity.

pip install pyenvector

import pyenvector as ev

---

🔍 Vector Search

1. Initialize enVector

Set up the connection to the enVector server and configure your key material.

ev.init(
    host="localhost",
    port=50050,
    key_path="./keys",
    key_id="quickstart_key",
)

---

2. Create Index

Create an index before inserting data. Define the index name and vector dimensionality.

index = ev.create_index("quickstart_index", dim=512)

---

3. Insert Data

Insert vectors into the index. The snippet below generates random 512-dimensional vectors with normalization and metadata.

import numpy as np

# Function to generate normalized random vectors
def generate_random_vector(dim):
    if dim < 32 or dim > 4096:
        raise ValueError(f"Invalid dimension: {dim}.")

    vec = np.random.uniform(-1.0, 1.0, dim)
    norm = np.linalg.norm(vec)

    if norm > 0:
        vec = vec / norm

    return vec

# Prepare sample data
num_data = 10
db_vectors = [generate_random_vector(512) for _ in range(num_data)]
db_metadata = [f"data_{i+1}" for i in range(num_data)]

# Insert into index
index.insert(db_vectors, metadata=db_metadata)

---

4. Encrypted Similarity Search

Query the index and retrieve encrypted similarity scores that are decrypted on the client side.

query = db_vectors[1]
top_k = 2

result = index.search(query, top_k=top_k, output_fields=["metadata"])
print(result)

---

🧹 Clean Up

Remove test data and release keys when you are done.

ev.drop_index("quickstart_index")
ev.delete_key("quickstart_key")

---

API Notes

• Indexer.insert_data_row uses a single cluster_id (int). A single-item list is accepted for backward compatibility.