AI Agent
Let an AI see your phone's screen and operate it for you
AI Agent exposes a safe, watched set of actions on the phone through Android's Accessibility service — tap, swipe, type, open app, read the UI tree, screenshot. A free controller on your PC (with a plain-rules planner by default, or Ollama / Claude / OpenAI / Gemini as optional providers) sends one action at a time, re-reading the screen after each step so it never fires blindly. Everything is token-authenticated and LAN-only, with an Emergency Stop and a full action log on the phone.
Also needs: Free controller (Node.js) on your PC — works with no paid API by default.
What it does
- AI reads the screen and taps / types for you
- Observe → plan → one action → verify loop (no blind clicks)
- Free by default (rule-based) — Ollama / Claude / OpenAI / Gemini optional
- Web dashboard with live screen and command box
- Strict action allow-list — no arbitrary shell
- Emergency Stop and full action log on the phone
What you need
- The free controller running on a PC on the same Wi-Fi
- Accessibility service enabled for the app (one toggle)
- Same shared token on both sides
Install
- 1.Run the controller on your PC (Node.js) — it prints an address and token.
- 2.Install this APK and open it.
- 3.Turn on the app's Accessibility service, paste the address and token, connect.
How to use
- 1
Enable accessibility
Open AI Agent → Open accessibility settings → turn on "AI Agent control". Without this the agent can neither read nor tap.
- 2
Connect to the controller
Start the controller on your PC, then in the app paste its ws:// address and token and press Connect.
- 3
Give a command
In the PC dashboard type something like "Open Settings". The agent looks at the screen, does one step, checks the result, and continues.
Safety
- 🛡Only a fixed allow-list of actions can run — there is no remote shell.
- 🛡LAN-only and token-protected; Emergency Stop halts everything.
- 🛡Every action is listed in the app, so nothing is hidden.