YourHand: an open-source bridge that lets one AI chat control multiple Windows PCs

One conversation, several machines

An open-source project called YourHand gives AI chat clients the ability to operate real Windows computers, and not just one of them. According to a post on dev.to, the tool is a Windows control layer that connects supported AI clients to one or more PCs, so a single conversation can open applications, read the screen, click, type, manage files, run commands and coordinate work across several machines, rather than pushing everything through a browser.

The author's stated motivation is personal: daily work spread across multiple Windows machines, each with its own applications, files and sessions, while an ordinary assistant can only advise and the human still jumps between computers executing each step by hand.

A layered architecture

YourHand stacks several layers: the AI client at the top, a YourHand MCP plugin, a control plane, a capability router, and a lightweight agent running on each Windows machine, with the OS, browser, UI, files and commands at the bottom. The control plane manages account identity, device ownership, sharing and request routing. Each agent exposes capabilities such as window and application control, keyboard and mouse input, screen observation, file access, command execution, browser control, and device health telemetry, and it runs quietly in the background so the user does not need to keep a terminal open for the machine to stay controllable.

Routing beats blind clicking

A central design decision, the developer writes, is that desktop automation should not depend on a single technique. A click can travel several possible paths: direct OS or application APIs, the Chrome DevTools Protocol when the target is a web page, semantic Windows UI Automation, pixel-based interaction, foreground input as a fallback, or a vision-assisted fallback when the UI cannot be addressed semantically. Instead of treating a button press as one primitive, YourHand routes each task to the most fitting capability, which the author argues makes it far more dependable than approaches built entirely on screenshots and blind cursor movement.

Multi-device and sharing by design

Many automation tools assume one session, one machine and one active task, and that was not enough for this project. Devices are treated as explicit resources tied to a user's account, so a conversation can target different computers by name and the control plane forwards each request to the correct agent. Sharing is handled at the same level: one installed agent can stay in place while access is granted to another authorized account, so a teammate does not need a second, competing agent on the same device.

Authentication and telemetry

Authentication uses Google sign-in and OAuth-based authorization for the YourHand account, with devices paired to that account. The design deliberately separates concerns: the AI client is authorized to call YourHand, YourHand decides which devices the account may use, and the Windows agent only accepts work routed through the control plane. The point, per the post, is to spare users from pasting AI-provider passwords or long-lived manual tokens into a desktop agent.

Observability mattered more than expected, the developer notes, because desktop automation fails in messy ways: commands that time out even though the action happened, windows vanishing between discovery and input, browser tabs dying while the browser lives on, stale sockets after a device reconnects. YourHand records success and failure, execution method, latency, retries and device health per call, making it possible to distinguish routing, UI, transport and application failures instead of blaming every incident on the model clicking the wrong thing.

Beta status

The project is in beta and free during that period, with a community repository on GitHub and a product website. The author positions it as more than a demo, with the architecture aimed at multiple devices and authorized users, account isolation, browser and native UI control, background execution, recovery after restarts, telemetry and controlled sharing. Current testing focuses on reliability, speed (preferring direct APIs and semantic automation before falling back to screenshots), multi-device workflows, and onboarding for people who have never configured an MCP server or a Windows automation agent.

Why it matters

Most desktop-agent efforts concentrate on a single machine and often lean on screenshot-driven interaction. YourHand's bet is that the useful unit of work is a fleet of real Windows PCs addressed from one continuing conversation tied to an account, a model that fits IT and engineering teams whose work is genuinely spread across computers. The layered authorization and per-call telemetry point toward automations that can be audited rather than blindly trusted, and publishing the code makes that trust model inspectable. The developer closes on the question that will decide adoption: what would it take to let an AI chat work across the computers you actually use?