Skip to content
View topmount121's full-sized avatar

Block or report topmount121

Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
topmount121/README.md

👋 About me

name:       World of Dots
role:       Senior Engineer
focus:      on-device AI · Android agents · local-first runtimes
working_on: 🚀 touchpilot — safe, observable phone control by AI
mission:    keep the model next to the screen it operates
philosophy: typed tools, explicit risk, audit by construction
open_to:    collaborators on local agents, mobile AI, small models

I spend most of my time on the boundary between large models and real devices — how to make an agent useful on the phone you actually carry, without leaking your screen to someone else's cloud.


🧰 Tech I reach for

skills


🚀 Featured project — touchpilot/touchpilot

🌟 Local-first Android AI agent runtime for safe, observable phone control.
Inspired by OpenClaw and PicoClaw — scoped around one hard problem first:
letting an AI agent operate an Android device through explicit, permissioned, inspectable tools.

✨ Why it's worth a star

📱 Native Android, not a desktop companion Runs on the device, observes the device, acts on the device.
🛡️ Typed tools with explicit risk levels Low-risk reads run silently. Medium/high-risk taps, typing, and app launches require user approval in chat.
👁️ Semantic UI via AccessibilityService The agent sees structured nodes, not screenshots — plans are grounded and cheap to verify.
🏠 Local-first routing Ships with a LiteRT command-routing model so simple actions never leave the device.
☁️ Cloud is a fallback, not the default OpenAI-compatible endpoints are experimental — opt-in, not opt-out.
📜 Auditable by construction Every model decision and tool execution is recorded in inspectable local logs.
🔐 Keystore-backed secrets API keys are encrypted with an Android Keystore-backed key before being stored.
🧩 MIT, Kotlin, contributor-friendly Issues are scoped. The runtime is small enough to read in a sitting.

🎯 Who should care

  • 🔬 Researchers comparing on-device vs. cloud agent loops on real hardware
  • 📲 Android engineers who want a reference for AccessibilityService-driven control
  • 🧠 Anyone building agent UX where approvals, audit logs, and risk levels are first-class

⚡ Try it in two minutes

git clone https://github.com/touchpilot/touchpilot
cd touchpilot
./gradlew assembleDebug
# Install the APK on an emulator (LDPlayer works) or a real device

👉 github.com/touchpilot/touchpilot
If you star one repo this week, let it be this one. 🌟


🤝 Let's build together

If you're working on local agents, Android automation, or small-model routing — open an issue on touchpilot or follow along. The roadmap has room.

Pinned Loading

  1. topmount121 topmount121 Public

    8