taOS provides a self-hosted AI agent operating system designed for users who prioritize data sovereignty and offline capabilities. It transforms diverse consumer hardware into a distributed AI compute cluster, allowing agents' memory, conversations, and files to remain on user-owned machines, with cloud integration as an optional choice. This platform is ideal for developers, researchers, and power users seeking a flexible, framework-agnostic environment for AI agents.

How It Works

taOS operates on a framework-agnostic principle, managing the entire agent lifecycle—memory, files, communication, and model access—independently of specific AI frameworks like LangChain or OpenClaw. This unique approach ensures agents retain their complete history and configurations when switching execution engines, eliminating migration complexities. Core components include taOSmd, a sophisticated framework-agnostic memory system leveraging temporal knowledge graphs and hybrid vector search; a distributed compute cluster that aggregates heterogeneous hardware; and a comprehensive web desktop environment providing a unified user interface. Backend-driven discovery ensures dynamic capability allocation based on available hardware.

Quick Start & Requirements

The controller can be installed on Debian/Ubuntu/Fedora/Arch/Alpine/macOS via a single-line curl command: curl -fsSL https://raw.githubusercontent.com/jaylfc/taOS/master/scripts/install-server.sh | sudo bash. Worker installation follows a similar pattern for Linux/macOS and PowerShell for Windows. The system supports a wide range of hardware, including Apple Silicon, NVIDIA/AMD GPUs, Rockchip NPUs, Raspberry Pi, and Android phones, with minimal platform overhead (~345 MB RAM). Official website: taos.my.

Highlighted Details

• taOSmd Memory System: Achieves 97.0% end-to-end Judge accuracy on LongMemEval-S, featuring a temporal knowledge graph, hybrid semantic+keyword vector search, and LLM-assisted query expansion.
• Distributed Compute Cluster: Seamlessly integrates diverse hardware (desktops, SBCs, phones) into a unified AI compute mesh.
• Web Desktop Environment: A full browser-based OS with a window manager, dock, launchpad, 36 bundled apps, and PWA support for mobile/tablet devices.
• Framework Agnosticism: Agents maintain persistent memory and connections across different AI frameworks without data loss or reconfiguration.
• TurboQuant KV Cache Compression: Enables significantly larger context windows (up to 768K tokens) on consumer GPUs by compressing KV caches.
• Exo Distributed Inference: An experimental feature for pooling VRAM across multiple machines to run models exceeding single-device capacity, primarily for Mac and GPU users.

Maintenance & Community

The project is in Beta (as of 2026-06-02) and actively developed, with CI running on every push. Community interaction is facilitated through GitHub Discussions. Notable community contributors are recognized, indicating active engagement and feedback.

Licensing & Compatibility

taOS is released under the taOS Sustainable Use License v0.1, which is source-available but not strictly open-source. It permits free use, modification, and self-hosting for personal and internal business purposes. Commercial use, such as selling the software, hosting it as a paid service, or incorporating it into a monetized product, requires a separate commercial license from jaylfc.

Limitations & Caveats

As beta software, taOS may exhibit rough edges, particularly in agent management and worker connection flows. Some application and framework installation manifests require further testing on diverse hardware. Running taOS within nested container environments like Proxmox necessitates specific privileged configurations. The Exo integration is experimental and currently lacks support for ARM/Rockchip hardware for model splitting. Certain advanced features, like cluster-wide scheduler aggregation and inline inter-framework delegation, are planned for future releases.