An AI agent that physically operates a phone — the way you do.

English docs · Demo video · Assembly video

中文文档 · 演示视频 · 装配视频

PhysiClaw watches a phone's screen with a camera and taps it with a stylus, working the phone the way a person would. No APIs, no OAuth, no ADB cables, nothing installed on the phone — just unlock it, set it on the desk, and let the agent work.

It's built for the everyday errands that pile up — ordering takeout, shopping for groceries, booking a ride, paying a bill, replying to a message. Anything you can do by hand on your phone, PhysiClaw can do for you.

The apps that run your daily life are closed off: most expose no public API, and simulated input — desktop automation or Android's ADB — leaves software fingerprints that anti-bot systems flag. So PhysiClaw treats the screen itself as the API: a camera reads it, a stylus performs the gestures. To the phone it's indistinguishable from a real finger — nothing to detect — and it reaches virtually any app reliably.

The trade-off is speed: a few seconds per action, in exchange for universality and reliability.

PhysiClaw has its own dedicated phone running its own WhatsApp account. Add it as a contact and chat with it like a real person — tell it what you need in plain language:

Order a latte on DoorDash.
What's the weather tomorrow?
Buy milk and eggs on Instacart.

Its runtime loops continuously, waking the agent when a scheduled task is due or the screen lights up with a new message. The agent unlocks the phone, reads your message, and does the task — pausing for confirmation when it matters, like a payment — then replies with the result, saves its memory, and exits until it's next woken.

You'll need the PhysiClaw hardware — the arm, stylus, and camera — plus an iPhone for it to operate.

The physiclaw CLI runs on macOS, Windows, and Linux:

# Install the CLI (uv + Python 3.12 + physiclaw)
curl -fsSL https://physiclaw.ai/install.sh | bash
# Windows: iwr -useb https://physiclaw.ai/install.ps1 | iex

physiclaw models key <provider>        # add your LLM API key
physiclaw models use <provider/model>  # pick the model the agent runs on
physiclaw doctor                       # check your environment
physiclaw                              # start the server — opens the hardware-setup wizard, then runs the agent

Run make firmware to build the FluidNC bundle used by physiclaw flash. See firmware/README.md for options, testing, and publishing.

MIT — both the CAD-as-code hardware and the agent software.