agents-io

agents-io / PokeClaw

Public

PokeClaw (PocketClaw) — first on-device AI that controls your Android phone. Gemma 4, no cloud, no API key. Poke is short for Pocket.

accessibility ai-agent android-automation gemma4 litert
38
1
100% credibility
Found Apr 06, 2026 at 38 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Kotlin
AI Summary

PokeClaw is an Android app using on-device AI to autonomously control the phone screen by reading UI elements and performing actions like tapping, swiping, typing, and opening apps.

How It Works

1
📱 Discover PokeClaw

You hear about this free app that lets AI control your phone without internet or bills, and download it from the project page.

2
📥 Install the app

Tap to install on your Android phone and allow screen reading permission when asked.

3
🧠 Download AI brain

The app grabs its thinking power (a few GB file) automatically on first use—no setup needed.

4
⚙️ Ready to go

Everything sets up itself; a floating button appears to show it's waiting for your command.

5
💬 Give a command

Type something simple like 'open camera' or 'reply hi to mom' and watch the magic.

🎉 AI takes over

Your phone reads the screen, taps and types on its own, finishing the job perfectly every time.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 38 to 38 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is PokeClaw?

PokeClaw is a Kotlin Android app packing the first on-device Gemma AI that controls your phone—no cloud, no API key, no internet. Tell it to open apps, tap screens, swipe, type messages, or auto-reply, and it reads the UI tree, picks tools, and acts in a local loop. Solves phone automation for offline users tired of cloud dependencies.

Why is it gaining traction?

Zero billing and full privacy hook devs: Gemma 4's tool calling runs smoothly on-device via LiteRT-LM, delivering tasks like navigating apps or sending context-aware replies without phoning home. Demos on budget phones (45s warmup on CPU) impress, especially versus API-locked alternatives. Early buzz from its 2-day prototype status draws AI tinkerers.

Who should use this?

Android devs prototyping local agents, mobile AI researchers benchmarking Gemma on phones, or power users automating daily tasks like messaging without cloud controls. Targets flagships (Pixel 8/9, Snapdragon 8 Gen 2/3, Dimensity 9300) with 8GB+ RAM for reliable on-device performance.

Verdict

Grab it for Gemma phone experiments—38 stars and 1.0% credibility scream raw prototype, with bugs and CPU slowness on low-end hardware. Solid proof-of-concept; PRs could make it production-ready fast.

(178 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.