# MemMe

A local, Rust and SQLite memory engine for AI pet and companion applications, keeping owner facts, per-relationship memories and shared history on the device and persisting them across restarts, model changes and network loss.

MemMe is a private, durable memory runtime meant to be embedded in AI-pet and AI-companion products rather than used as a standalone application. It is designed to help an AI pet remember its owner, keep each pet's relationship with its owner separate from other pets, preserve shared experiences, and continue working after restarts, model swaps or a loss of network connectivity, distinguishing owner memory such as stable facts and boundaries from relationship memory such as names and shared history.

A bundled demo runs entirely with local inference, requiring macOS or Linux, Rust, Python 3 and curl, and needs internet only on first run to download a local ONNX embedding model and the VexDB-Lite storage extension; it writes memories for one owner and two pets, restarts the server, and recalls them again to demonstrate restart persistence and per-pet isolation.

The engine's Rust core is intended to be embedded in other companion products rather than run alone. It is licensed under Apache-2.0.

Website: https://vibeinging.github.io/MemMe/