Let your AI agent control any physical device - Phones, Tablets, PCs, and more.
Early Stage Project — AgentKVM is still in very early development. APIs, commands, and behavior may change. Issues and contributions are welcome!
AgentKVM is a CLI tool that lets AI agents see and control physical devices through KVM hardware.
Supported devices:
- NanoKVM-USB — currently supported
Planned:
AgentKVM bridges the gap between AI and the physical world. An AI agent takes screenshots to observe a device's screen, then sends mouse clicks, keyboard input, and scroll events to interact with it - enabling true "computer use" on real hardware.
- See — Capture the device's screen via HDMI →
agentkvm screenshot - Think — AI analyzes the screenshot to understand the UI
- Act — Send mouse/keyboard input via USB HID →
agentkvm mouse click,agentkvm type,agentkvm key
Copy the prompt below and send it to your AI agent - it will guide you through the entire setup process. Of course, if you're a human, you can also follow the Quick Start Guide yourself.
Help me install and configure AgentKVM (https://github.com/iamtwz/agentkvm).
1. First, read the Quick Start guide: https://raw.githubusercontent.com/iamtwz/agentkvm/main/QuickStart.md
2. Then follow the steps to install AgentKVM and set up my device.
AgentKVM provides an Agent Skill that lets AI agents use it directly. Install with either:
# Via Agent Skills
npx skills add iamtwz/agentkvm
# Via OpenClaw (ClawHub)
npx clawhub@latest install agentkvmOnce installed, your AI agent will automatically know how to take screenshots, click, type, and scroll on your connected devices.
git clone https://github.com/iamtwz/agentkvm.git
cd agentkvm
pnpm install
pnpm dev -- list # Run in development mode
pnpm test # Run tests (204 tests)
pnpm typecheck # TypeScript type check
pnpm build # Build for productionMIT