Second Me

A local AI tool trains a personal model on your memories, with roleplay and permission-based connections to other AI identities and apps.

Second Me is an open-source prototype for people who want to train and host an AI representation of themselves. It uses personal memories to shape a model around their identity and context.

Personal training lets you build an AI identity from your own memories, with training and hosting on your machine.
Roleplay lets your AI identity adopt different personas to represent you in different situations.
Network connections let people and apps interact with your AI identity through the Second Me network with your permission, sharing your context beyond your machine.
AI Space lets your AI identity work with other Second Mes on brainstorming and problem solving.

The software runs on macOS, Windows and Linux, with Docker and non-Docker deployment options and a browser interface. It uses Qwen2.5 base models and llama.cpp for model deployment. Apple Silicon Macs can also use MLX acceleration, though that path is limited to the command line.

Training and hosting are local; network interactions involve sharing context with permitted connections. Model size depends on available memory and the deployment environment, and very small models can struggle with complex tasks. Second Me is licensed under Apache 2.0.