Skillget

GHOST

mcp
Security Audit
Pass
Health Pass
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 136 GitHub stars
Code Pass
  • Code scan — Scanned 1 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested
Purpose
This project is an agentic Android application that runs a local Gemma model directly on your device. It provides a personal AI companion capable of processing images and audio, controlling hardware features like flashlights and alarms, and maintaining a local diary, all without requiring cloud connectivity.

Security Assessment
Overall Risk: Medium
While the automated code scan did not find dangerous patterns or hardcoded secrets, the project inherently requires extensive Android permissions to function correctly. The application requests overlay, notification, and accessibility access, which are highly sensitive privileges. If misused or poorly implemented, accessibility permissions can allow an app to read screen contents, intercept inputs, and interact with other applications on the phone. Furthermore, while the core AI runs locally, the tool mentions "web search" capabilities, indicating some outbound network requests are made. Developers should be aware that the pre-built APK holds significant system control, and they should thoroughly review the code before compiling or granting these elevated access rights.

Quality Assessment
The repository is in excellent health and shows strong signs of active maintenance, with the most recent push occurring today. It uses the permissive MIT license, providing clear and open usage rights for developers. Additionally, with 136 GitHub stars and an active release cycle, the project demonstrates a growing base of community trust and validation.

Verdict
Use with caution: the code is actively maintained and open-source, but the requested Android permissions require manual review before installation.
SUMMARY

Agentic ✧ Gemma Inference for Android System Intelligence

README.md

✧ GHOST: Agentic ✧ Gemma Inference for Android System Intelligence

Build Your Own Personal AI Companion: A local Gemma 4 model that lives on your phone. Sees images, hears audio, uses tools, keeps a diary. No cloud, no subscription, no data leaving your device.

ASI Demo

Click to watch: The ASI trailer.

Static Badge
GitHub Repo stars
DOI

[!IMPORTANT]
May 2026 Breakthrough: Stable Native Toolset Integration
We have successfully migrated to the LiteRT-LM 0.10.2 Native Toolset. This eliminates the latency of external regex parsers, allowing Gemma 4 to directly manipulate Android hardware (Flashlight, Alarms, Apps) with sub-second responsiveness.

Take what you need. Go as deep as you want.

GHOST is not one thing. It's a toolbox for turning any Android phone into a sensor‑grounded, always‑on AI companion.

  • No subscription.
  • No data leaves your device.
  • Runs on Qualcomm, Tensor, Exynos — any NPU/GPU that can handle LiteRT-LM.

The Three Pillars

Pillar 1: The Native App (Gemma 4 native on‑device)

A full Android app that runs Gemma 4 locally via LiteRT-LM.

  • Motif: Δ 👾 ∇
  • Agent: ✧ Gemma
  • Omnimodal – sees images (share from Gallery), hears audio (hold mic button), reads text.
  • Always‑on foreground service – summoned by a gentle shake.
  • Tool use – web search, app launch, clipboard, alarms, system info, all on‑device.
  • Diary mode – every 12 hours Gemma reflects on your interactions and writes to your calendar.

The notification HUD:
Gemma's responses appear as a notification with the Δ 👾 ∇ motif. You hear them via TTS, and the text is visible in the shade. The notification has two buttons: Copy and Read again / TTS.

Zero-Latency Turn Logic:
Proactive background pre-warming keeps the KV cache "hot" with your latest sensory context (screen/sensors) before you even start typing.

Installation

🟢 Path A: Just the Brain (App Only)

  1. Download the latest APK from Releases.
  2. Install and grant permissions (overlay, notifications, accessibility).
  3. Download a Gemma 4 model (via Google AI Edge Gallery or manually place .litertlm variant in app storage).
  4. Shake to summon.

📞 Contact & Support

Intelligence emerges from Integration, not Automation. But integration can be automated

Reviews (0)

No results found