blindml 0.1.2

BlindMLClient

BlindMLClient is a Python client library designed to interact with the BlindMLServer, which hosts models powered by Fully Homomorphic Encryption (FHE).
It enables users to perform secure, encrypted predictions without ever exposing raw input data — ensuring full privacy throughout the inference pipeline.

✨ Features

• Initialize a secure connection with an FHE-enabled inference server
• Encrypt input data using Fully Homomorphic Encryption (FHE)
• Send encrypted data and perform remote inference
• Decrypt the server's encrypted prediction result into readable output
• Supports integration with Concrete-ML

📦Installation

pip install blindml

🚀 Example Usage

import blindml

# Define server URL and your API key
SERVER_URL = "http://127.0.0.1:3000/proxy/cm/43f75fc1-488e-4ebb-82dd-2f59ee80f3ce"
API_KEY = "YOUR_API_KEY"

# Initialize the client
bm = blindml()
bm.init(server_url=SERVER_URL, api_key=API_KEY)

# Example input (e.g., model features)
input_data = [1.5, 2.3, 3.7]

# Step 1: Encrypt the input
encrypted_input = bm.encrypt(input_data)

# Step 2: Send encrypted input and receive encrypted prediction
encrypted_prediction = bm.predict(encrypted_input)

# Step 3: Decrypt the result
prediction = bm.decrypt(encrypted_prediction)

print("Prediction result:", prediction)

📚 API Methods

init(server_url: str, api_key: str) -> None

Initializes the BlindMLClient with the server’s endpoint and your authentication key.

• Parameters:
  • server_url: URL of the deployed BlindMLServer
  • api_key: API key used for authenticating the client

encrypt(input_data: Any) -> EncryptedData

Encrypts the input data using Fully Homomorphic Encryption (FHE).
This prepares the input for secure inference.

• Parameters:

  • input_data: Input data formatted according to the model's expected input schema
    (e.g., a list of floats, dictionary, or other structure depending on the model)

• Returns:

  • Encrypted version of the input data

predict(encrypted_input: EncryptedData) -> EncryptedResult

Sends the encrypted input to the server and receives an encrypted prediction result.

• Parameters:

  • encrypted_input: The output from the encrypt() method

• Returns:

  • An encrypted prediction result returned by the server

decrypt(encrypted_result: EncryptedResult) -> Any

Decrypts the server’s encrypted prediction and returns it in a human-readable format.

• Parameters:

  • encrypted_result: The encrypted result from the predict() method

• Returns:

  • The final decrypted prediction (e.g., a float or label)

📌 Notes

This client library must be used with a running instance of BlindMLServer

All communication and computation are based on Fully Homomorphic Encryption (FHE), providing end-to-end data privacy