Tapilotbeta

Powered by Midscene

Your Android phone.
Your words. No computer.

Tell Tapilot what you want to do. It opens apps, taps and types on your Android phone. Start with a small task, watch it happen, then check the result.

Android 11+ · ARM64 · Download: 0.1.9 Beta

GitHub download · v0.1.9 ↗

Released November 4, 2026. Verified on an Android phone; signed for compatible beta upgrades.

YouTube · EnglishOpen video ↗

Playback trouble? Switch source or open the video.

78 SECONDS / TAPILOT IN ACTION

One phone. Four ways to see it work.

Product film with device footage and edited report views. Features depend on the installed build.

01 / TAPILOT

From your words to the screen.

Describe the task, add a requirement if needed, and inspect what happened on your phone.

01

Let your phone do the tapping

Ask it to open an app, find something on screen or enter text. Everyday use needs no computer, USB cable or root.

02

Keep the task moving

Add a requirement while a task runs, stop it, or retry explicitly. Interrupted tasks do not replay automatically after a restart.

03

Check the actual result

Each phone operation gets a separate screenshot check. Open a replayable report to inspect screenshots, actions and errors.

04

Connect your computer

A compatible build can pair with the Android Use plugin over Wi-Fi. Keep the phone unlocked and give one controller exclusive access.

02 / GET STARTED

Try your first task.

  1. 1

    Connect your model

    Configure an HTTPS vision-model API, or use the available ChatGPT connection option. HTTP is also supported for self-hosted models on trusted networks. See Getting started for setup.

  2. 2

    Enable device control

    Enable Tapilot in Android Accessibility settings. Choose automatic execution, confirmation for critical operations, or confirmation for every operation.

  3. 3

    Calculate 12 × 8

    Send “Open Calculator and calculate 12 × 8.” Keep the phone unlocked, watch the steps and check that the screen shows 96.

VOICE & CONVERSATION

Speak, then keep the conversation going

Configure Volcengine or Deepgram for voice input. Add voice playback for spoken replies, or an optional polishing model to tidy up transcripts. If polishing fails, the original text is kept.

IN YOUR HANDS

Your providers. Clear data paths.

The agent runs on your phone; model inference uses your configured model service. Keys are encrypted with Android Keystore. Screenshots and task context go to the selected model, and optional speech services receive audio or reply text.

Privacy & permissions →

Give your next task to your phone.

Download Android beta ↗

Tapilot operates on your main screen: keep your phone unlocked and avoid using it during a task. This beta can make mistakes. A supported vision model is required; provider fees may apply.