jaylfc/taOS

What it solves

taOS is a self-hosted AI agent platform designed to give users full sovereignty over their AI experience. It eliminates reliance on cloud providers by allowing users to run agents, memory, and models on their own hardware—ranging from old laptops and Raspberry Pis to gaming PCs—while keeping data fully offline by default.

How it works

The system operates as a distributed compute cluster where a central controller manages multiple worker nodes (various devices on a local network). It features a browser-based web desktop environment that acts as the primary interface for managing agents, apps, and system health.

Key technical components include:

• taOSmd: A framework-agnostic memory system using a temporal knowledge graph and hybrid vector search to provide high-accuracy long-term memory for agents.
• Distributed Workers: Cross-platform worker apps (Linux, macOS, Windows, Android) that connect to the controller to provide CPU, GPU, or NPU compute for inference and embeddings.
• Framework-Agnostic Layer: taOS manages the agent lifecycle (memory, files, keys) outside the specific agent framework, allowing users to switch between frameworks (e.g., SmolAgents, LangChain) without losing data.

Who it’s for

• Privacy-conscious users who want their AI conversations and memory to stay on their own hardware.
• Hobbyists and developers who want to build a distributed AI cluster from spare hardware.
• Users seeking a unified, self-hosted interface to manage multiple AI agents and models across different hardware architectures.

Highlights

• Web Desktop OS: A full browser-based environment with a window manager, dock, and 43 bundled apps (including a code editor, terminal, and image studio).
• Hardware Versatility: Supports Apple Silicon (MLX), NVIDIA, AMD, Rockchip NPU, and Raspberry Pi.
• Sovereignty by Default: Data is stored locally; cloud models are optional and opt-in.
• High-Accuracy Memory: The taOSmd system achieves high end-to-end judge accuracy on LongMemEval-S benchmarks.
• Multi-Framework Support: Ability to run agents from different frameworks (e.g., OpenClaw, Hermes, Langroid) in a single shared chat channel.