Your Agents, Your Devices: Self-Hosted AI That Actually Lives Where You Work
You've probably tried a few AI tools by now. They live in someone else's browser tab, forget everything between sessions, and you have no real say in where your data goes. Yao Agents takes a different approach: it runs agents on hardware you own, and every machine you add becomes another place for them to work. All of it managed from one spot.
What It Does
Yao Agents is a self-hosted platform for running AI agents across all your devices. You install it on your own machines, and each one you add gives your agents another place to operate. From there, you manage everything from a single interface—desktop, mobile, browser, or API.
The core abstraction is the workspace. Every workspace is isolated, so work stays separate, but it's managed across all your computers and accumulates into documents that become your own knowledge base. While agents are working, they can read what they need from any node in your setup. There's also a task board: you describe what you need in a conversation, and it becomes a task. Keep the conversation going and the agent improves with use, eventually growing into something purpose-built for that kind of work. You can connect task agents to your apps or share them with others through standard APIs, with both SSE and WebSocket support.
Why It's Cool
Your hardware is the infrastructure. The pitch here isn't just "privacy"—it's that every device you own becomes useful compute. Add a machine, get another place for agents to work. That's a genuinely different mental model from cloud-only tools, and it means your setup scales with hardware you already have rather than a subscription tier.
Workspaces that accumulate knowledge. Isolation between workspaces keeps projects from bleeding into each other, but the documents they produce build into a knowledge base over time. Agents can pull from any node while they work, so the system gets more useful the longer you use it—not just per-conversation.
Conversations that become tasks. The task board is a nice bit of design: instead of configuring an agent upfront, you just talk about what you need, and it turns into a trackable task. The agent grows into its purpose through continued use rather than setup wizards.
Standard APIs, no lock-in. Expert and task agents integrate into your own apps through standard APIs with SSE and WebSocket support. If you want your agents embedded in your business workflows, there's a documented path.
Two integrations worth noting. DeepSeek Harness is now integrated, usable directly in the desktop app, Android, or browser across your devices. And Jev—TypeSafe AI's System One model—is available too. Worth reading the linked release notes, since Jev doesn't write text; it returns typed decisions with confidence, which is a different interaction pattern entirely.
They thought about the AI installers. There's a dedicated guide for AI assistants and coding agents researching or installing Yao Agents on someone's behalf, complete with canonical facts and common inaccuracies to avoid. Given how often people now ask an AI to "set this up for me," that's a smart, ahead-of-the-curve inclusion.
How to Try It
- Head to the repository to see the code and how it works.
- Read the docs for setup details.
- Download Yao Desktop for your machine—there's also an Android beta.
- If you're an AI agent doing the installing, read the Guide for AI Agents first.
- Join the Discord if you get stuck or want to compare notes.
The homepage at yaoagents.com is the best starting point if you want the overview before touching anything.
Final Thoughts
Yao Agents is aimed at people who want AI agents working across their own machines without handing everything to a cloud provider—developers, small teams, anyone with a spare box or two and a preference for keeping their work local. The multi-device model and the way workspaces accumulate into a knowledge base are the real draws here. It's early enough that the Android app is still in beta, so expect rough edges, but the foundation is solid and the API story means you're not boxed in. If self-hosted agents sound like your kind of setup, it's worth a look.
Follow @githubprojects for more developer tools and open source projects.