It uses offline models to store memory on your device, remembers what you share, and helps you find it later. Memory is saved as Markdown and organized into a clear, connected graph for you and your agents to navigate.
Available for macOS 14+ with Apple Silicon (M1 or better).
Everything stays local, and you can start a memory-aware chat session instantly from the terminal. The experience is powered by open weights models fine-tuned specifically for memory.
Interactive graph that lets you explore your memories and agent skills in a clean, connected view.
Cataloged in Registry, these models are trained with RL for memory operations on Markdown files so the agent can maintain and refine persistent knowledge across sessions, while optional LoRA memory extensions let users and organizations bring their own data and shape the model's personality.
Powered by Tilekit
Explore the design of our Rust based Modelfile SDK for private, cross-platform model customization and access.