Second Me is an open-source project enabling users to train and deploy a personalized AI "self" that preserves their identity and context. It targets tech enthusiasts and AI professionals seeking to amplify their capabilities and control their digital presence, offering a decentralized network for AI selves to interact and a platform for building AI-native applications.

How It Works

The core of Second Me is its AI-Native Memory system, utilizing Hierarchical Memory Modeling (HMM) and a Me-Alignment Algorithm. This approach allows the AI to capture and reflect the user's identity and context authentically. The system is designed for local training and hosting, ensuring user privacy and control, while also enabling scalability through a decentralized network where AI selves can connect and share context with explicit user permission.

Quick Start & Requirements

• Docker Setup: make docker-up (Requires Docker and Docker Compose). Allocate at least 8GB RAM for Docker Desktop.
• Integrated Setup: make setup then make restart (Requires Python 3.12+, Node.js 23+, basic build tools). Offers better performance, especially on Mac/Linux.
• MLX Acceleration: Available for Mac M-series users (CLI-only) for running larger models.
• Documentation: FAQ, User Tutorial

Highlighted Details

• AI-Native Memory system with Hierarchical Memory Modeling (HMM) and Me-Alignment Algorithm.
• Decentralized network for AI self interaction and collaboration.
• Supports role-playing and AI Space for collaborative problem-solving.
• Emphasis on 100% privacy and local data control.

Maintenance & Community

• Active community channels: Discord, Reddit, X.
• Roadmap and development details are available via community channels and contribution guides.

Licensing & Compatibility

• Licensed under the Apache License 2.0.
• Permissive license suitable for commercial use and integration into closed-source projects.

Limitations & Caveats

• Windows support for integrated setup is not yet tested. Docker setup on Mac M-Series chips incurs a 25-30% performance overhead. Models below 0.5B parameters may yield unsatisfactory performance for complex tasks.