Second Me is an open-source AI platform developed by Mindverse that lets anyone create a locally trained, personalized AI representation of themselves. Unlike cloud-based personal AI assistants that store user data on centralized servers, Second Me runs entirely on the user's own hardware. It uses proprietary training techniques — Hierarchical Memory Modeling and the Me-Alignment Algorithm — to capture a user's communication style, preferences, and knowledge from their personal documents. The project has 15,100 GitHub stars and 1,200 forks, backed by a research paper titled 'AI-native Memory 2.0: Second Me' (arXiv 2503.08102). The Apache 2.0 license permits both personal and commercial use. ## AI-Native Memory Training At the core of Second Me is a fully automated post-training pipeline built on personal document records. The system uses Hierarchical Memory Modeling to organize memories at different abstraction levels — episodic details, semantic knowledge, and behavioral patterns — then applies the Me-Alignment Algorithm to ensure the resulting model genuinely reflects the user's identity rather than producing a generic assistant persona. This approach differs fundamentally from prompt-engineering-based personalization. Instead of instructing an off-the-shelf LLM to behave a certain way, Second Me fine-tunes a model on the user's actual data. ## Privacy-First Local Architecture All training and inference runs locally on the user's machine. No personal data is sent to external servers during the training process. This design enables use cases that are untenable with centralized AI services, such as processing sensitive personal documents or confidential business information. ## Decentralized AI Network Beyond personal use, Second Me supports deployment to a decentralized peer network. Users can selectively share their AI self with others under configurable permission controls, enabling collaborative scenarios where multiple Second Me instances interact in an 'AI Space'. ## Dual Interaction Modes Second Me offers two modes: Chat Mode for direct personalized conversations, and Bridge Mode where the AI acts as an intermediary between the user and external systems or other agents. ## Easy Docker Deployment The quick-start is three steps: clone the repository, run `make docker-up`, and open http://localhost:3000. Docker images are available for Windows, Linux, and macOS. The stack is primarily Python (76%) with a TypeScript frontend (19.5%).