Just found out about this awesome completely open-source platform. Build your own AI self that truly mirrors your will and memory, running 100% locally Their Github got 10.1K+ stars⭐️ in 3 weeks. @SecondMe_AI1 👏 github[.]com/mindverse/Second-Me 🧵 1/n What is Second Me: - Train Your AI Self; - Scale Your Intelligence; - Connect with a Network of AI entities - 100% Privacy and Control. - Think: fully local LLM + memory + personality + decentralized networking. Second Me captures personal context with Hierarchical Memory Modeling (HMM) across three memory layers. Me-Alignment refines those memories, producing a self-modeled AI identity tuned to each individual. This approach outperforms standard retrieval-based pipelines in personalization accuracy. It blends local security with global AI collaboration in a single platform. So how does it all work? ⚙️ Three-Layer Hierarchical Memory (HMM) Second Me splits user data into three layers: raw data (L0), text-based memory (L1), and a learned/parameterized memory (L2). L2 holds crucial patterns from your information so the AI can automatically recall context without scanning every file. 1. Users upload files, notes, or calendars into local folders. Second Me scans them and extracts important entities or keywords. It stores these as raw data (L0) plus text-based memory (L1). 2. Next, the system builds a learned/parameterized memory (L2) by applying a fine-tuning pipeline. It uses supervised fine-tuning (SFT) to teach the model how to answer queries in a personalized way, then direct preference optimization (DPO) to adjust answers according to the user’s unique priorities. Chain-of-Thought text can be included for deeper reasoning. 3. This training process runs entirely on the user’s machine. Nothing goes to external servers; all personal data and training stay local. 4. Once trained, Second Me can automatically provide relevant context or act as a “critic” when collaborating with external models. It captures the user’s perspective and supplies crucial details to make interactions more aligned with the user’s goals. 5. Each user’s Second Me can connect with other Second Mes in a decentralized fashion. This forms a network of AI identities exchanging ideas, but every instance remains under its owner’s full control.