Free and open source

The first agent for your iPhone.

Plenty of agents can write your code. This one can use your phone.

Nothing to install on your phone.No jailbreak, no profile, no app.
Works with the apps you already use.No integrations. If you can see it, it can use it.
Any model you choose.Every major provider, or your own endpoint.
Every step replayable.Watch exactly what it did, frame by frame.

Your iPhone stays exactly as it is.

No jailbreak, no profile, no app on the phone. It works through iPhone Mirroring on your Mac, so your iPhone stays exactly as it is.

Taps and keystrokes are injected in the background. The mirror never comes to the front, and your cursor never leaves your document.

No integrations to set up, no special support needed. If it is on the screen, it can read it, tap it and type into it.

Every run leaves a map of where it went and what worked. The same task took 11 steps the first time and 5 the next.

11steps, first run

5steps, next run

Use the model you already pay for, switch with one flag, or point it at one you host yourself.

iphone run --model deepseek:deepseek-v3 "…"

# .iphone/config.toml
[providers.myvllm]
base_url = "https://my.endpoint/v1"

Install, then hand it a task. Works with iPhone Mirroring on your Mac.

git clone REPO && cd iphone-agent && pip install -e '.[repl]'

iphone run "open Settings and tell me the iOS version"

Prefer a window? iphone serve opens the local app at 127.0.0.1:8765.

9:41

0apps installed 0profiles added 0settings changed

Your Mac · still yours

OCR · 18 words Vision · 6 elements

OpenAIAnthropicGeminiDeepSeekQwenKimiOllama

Ask it to…

Nothing on your phone, nothing hidden, nothing you can’t see afterwards.

Built on Apple’s own iPhone Mirroring.

The same feature you already use to see your phone on your Mac. Nothing bypassed, nothing patched, nothing added to the phone.
Runs on your Mac, with the model you choose.

You decide which provider sees your screen. Or point it at a model you host yourself and keep everything at home.
Every step recorded. Every run replayable.

See exactly what it tapped, typed and read, frame by frame, any time you like.

Works with any OpenAI-compatible provider, including one you host yourself.

Open source on GitHub. Star it, fork it, teach it your apps.

Skills are small files. Bring the apps you use every day, and the agent learns them for everyone.