PrivAgent

PrivAgent

3.3

PrivAgents is a privacy-first framework that uses autonomous agents and MCP servers to perform secure data processing with homomorphic encryption.

PrivAgents is designed to ensure complete confidentiality in data processing by leveraging Privacy-Enhancing Technologies (PETs), specifically Homomorphic Encryption (HE). It allows autonomous agents, whether on-device or cloud-based, to process encrypted data without accessing raw information. The framework is particularly useful in sensitive environments like healthcare, personalization, and finance. User data is encrypted on the device, processed by a secure MCP server, and returned in encrypted form for local decryption, ensuring an end-to-end secure pipeline.

Features

  • On-device or cloud-based agents for flexible compute and trust options.
  • Homomorphic encryption for secure data processing without exposing raw data.
  • Modular MCP server for encrypted similarity calculations.
  • End-to-end encrypted pipeline for secure data interaction.
  • Supports both Ollama and OpenAI agents for on-device and cloud-based intelligence.

Usage with Different Platforms

Ollama Agent (Local Execution)

bash
PYTHONPATH=. python agent/ollama_mcp_agent.py

OpenAI Agent (Cloud Execution)

bash
PYTHONPATH=. python agent/openai_mcp_agent.py