taOS: Run Self-Hosted AI Agents Across Your Hardware

Overview

taOS is a self-hosted AI agent platform built around taOSmd, its framework-agnostic memory system. It aims to keep agents’ memory, conversations, files, and messaging on hardware you own, while letting you choose whether to connect cloud models. Its web desktop brings agent management, chat, and system tools together; a distributed cluster can combine devices such as Raspberry Pi and Orange Pi boards, Apple Silicon Macs, PCs, and Android phones.

The platform separates agent lifecycle and data from the framework that executes an agent, so the framework can be changed without moving its memory or channel setup. taOS is aimed at people who want a local-first AI environment spanning modest single-board computers through GPU-equipped machines, rather than a hosted assistant or a framework-specific agent setup.

Key Features

• Framework-agnostic agent management, with agents’ memory, files, channels, and configuration managed outside the execution framework.
• taOSmd-backed memory with a temporal knowledge graph, hybrid search, and an append-only archive.
• A web desktop and mobile-friendly interface for chat, files, models, apps, and agent administration.
• Cluster workers that can contribute compute from supported Linux, macOS, Windows, and Android devices.
• A model catalog and live browser, with hardware-aware discovery for local inference backends.
• Messaging connectors for services including Telegram, Discord, Slack, Matrix, email, web chat, and webhooks.
• Agent deployment, shared workspaces, scheduled tasks, and a registry for MCP plugins and skills.

Use Cases

• Privacy-conscious individuals can run agents and retain their conversations and memory on a home server or single-board computer.
• Small teams can provide agents with shared channels, files, and project tools without relying on a hosted agent platform.
• Local-AI enthusiasts can pool a Mac, a gaming PC, and smaller devices into a cluster for different inference and support tasks.
• Developers experimenting with agent frameworks can switch an agent’s execution framework while keeping platform-managed memory and channels.

Getting Started

On a supported Debian, Ubuntu, Fedora, Arch, Alpine, or macOS controller, the README provides this installer:

curl -fsSL https://raw.githubusercontent.com/jaylfc/taOS/master/scripts/install-server.sh | sudo sh

Then open http://your-host:6969. See the README for platform-specific dependencies, worker setup, and development instructions.

Alternatives

• QwenPaw: QwenPaw focuses on a personal assistant connected to chat apps and extensible with skills, plugins, and MCP tools; taOS centers on an on-device platform with chat, files, and agents.
• ouroboros: Ouroboros is a persistent desktop or headless agent that works in external projects, while taOS provides a broader self-hosted platform for chat, files, and agents across devices.

Considerations

• The README labels taOS beta software. It says some agent-management, worker-connection, and model-routing flows are still being smoothed out, and warns that many catalog manifests have not been tested on real hardware.
• The project is active, with frequent commits and releases, but measured contributors are few and code contribution is highly concentrated. The measured period also recorded no issue closures, despite substantial merged pull-request activity.
• Python 3.11 through 3.14 is required. GPU or NPU hardware is not a general installation requirement, but model capability and performance depend on the devices and backends available; smaller devices may be limited to smaller models or CPU tasks.
• The AGPL-3.0 license should be reviewed before modifying or offering the software as a network service. The repository provides a self-hosted installation path, not a managed hosted service.