Phoneuse lets AI agents tap and type on an iPhone from Linux
A new MIT-licensed GitHub project mirrors an unmodified iPhone over AirPlay and sends Bluetooth input back, with a CLI and agent skill on top.
On 2026-11-14 the repository lxe/phoneuse showed 652 stars and 61 forks, collected in the seven days since it was created on 2026-11-07. The project, written in Python and released under the MIT license, is a working example of phoneuse iPhone agent control: it puts an unmodified iPhone screen in a window on a Linux PC and sends mouse and keyboard input back to the phone over Bluetooth HID. The README states that Claude Code, Codex and other agents can tap and type on the device, with the owner's permission.
What the source confirms
The input and output paths are split. Screen Mirroring (AirPlay) brings the iPhone display to the computer, and Bluetooth HID carries taps, swipes, scrolling and typing back out. No jailbreak and no cable are required.
Installation is a three-step process. On Linux you download a flatpak build and install it with flatpak install --user. On a Mac you need macOS 15 or later and the .dmg, and the source says the Mac app needs the bridge. Windows is currently from source only, with instructions in docs/WINDOWS.md. The second step is the Phone Use iPhone app from the App Store, and the third is opening the desktop app, which walks through connecting the keyboard and mouse, pairing the iPhone app and starting Screen Mirroring. The iPhone and the computer have to be on the same Wi-Fi network.
The bridge is a small LilyGO T-Display board that joins the Wi-Fi network and acts as the iPhone's keyboard and mouse, so the computer's own Bluetooth stays free. The Linux app can flash the board from a menu entry, and docs/BRIDGE.md covers other flashing routes. The project's own GitHub page links to phoneuse.tech, a download page and further documentation.
How a person or an agent takes control
Used directly, the app offers an Open iPhone button. Clicking taps, dragging swipes, the mouse wheel scrolls and the keyboard types.
For an agent, the app has a Connect Claude Code button. The README says Phone Use can ask the user on the iPhone before an agent takes control, and that the user can take over at any time. That approval step on the phone, rather than on the desktop, is the interesting design choice, because the person holding the device is the one who grants access.
What this likely changes for app builders
The obvious use is testing. This suggests a web team could keep a real iPhone in the loop while an agent drives it through a signup flow or a checkout screen, instead of relying only on a desktop browser in responsive mode. It also hints at a pattern where an agent skill owns a phone-based task end to end, with a human approving the handover. Both readings are inference; the source describes the mechanism, not specific workflows.
Limits and open questions
The project calls itself alpha software and asks users to report what breaks. The source does not state which iOS versions or iPhone models are supported, how much latency the mirror and input path add, or how the Wi-Fi, AirPlay and Bluetooth pieces behave on congested networks. It does not say what the bridge hardware costs or where to buy it, whether Android is planned, or what the agent skill can do beyond tapping and typing. Windows support is described only as from source for now. Anyone evaluating this for a team should treat the permission prompt as the start of a security review, not the end of one.
For teams building with AI
If the appeal here is giving a model hands, the same question comes up in web apps: where does the model sit, what can it touch, and who approves it. Draftship builds and hosts web apps with a built-in database and user sign-in, and its documentation on adding AI to your app covers that side of the design. Draftship has no connection to phoneuse.
Primary sources
- lxe/phoneuse: See and operate an unmodified iPhone from a Linux PC: AirPlay mirror in, Bluetooth HID mouse/keyboard out, plus a CLI and skill for agents. · GitHub trending AI repositories, Nov 14, 2026
This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.