# taOS

A self-hosted AI agent operating system that keeps an agent's memory, conversations and files on hardware the user owns, running offline by default across a cluster of consumer devices.

taOS is a self-hosted AI agent platform that runs on whatever hardware is available, from an old laptop or Raspberry Pi to a gaming PC, and can turn several such machines into a distributed compute cluster. It provides a web desktop environment with bundled and catalog apps, MCP plugins, a curated local model catalog covering language, vision, embedding, audio and image-generation models, and access to a large searchable set of models from Hugging Face, alongside agent deployment, training and system monitoring.

It is framework-agnostic: taOS itself owns an agent's memory, files, communication channels, model access and configuration, while the agent framework underneath, such as SmolAgents or LangChain, is a replaceable execution engine that can be swapped without losing history, chat-platform connections, trained adapters, files or API keys. A component called taOSmd provides a framework-agnostic memory system.

It supports Apple Silicon through MLX, NVIDIA, AMD, Rockchip NPU and Raspberry Pi hardware. The project describes itself as beta software; the install script, backend, API, memory system and multi-framework group chat work, while some desktop GUI flows and install manifests for its app, framework and model catalogs are still being smoothed out on real hardware.

Website: https://taOS.my