xjyzs/Operator-on-Android
xjyzs/Operator-on-Android has 35 GitHub stars and 6 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-12 13:03 UTC. Last push 2026-11-09.
On-Device Mobile Agent: Control your phone directly with Vision Language Models, untethered from any PC. 手机直接调用视觉大模型 ,无需连接电脑,让 AI 直接操作你的手机!
What you can do
- Virtual Screen Support: The AI operates on a background virtual screen without occupying your main display. You can chat, browse, or watch videos while the AI works in the background.
- Accessibility-based Layout Parsing: Leverages Android Accessibility Services to parse screen layouts, providing coordinates of UI elements and richer context for the AI.
- Token Consumption Tracking: Built-in stats to monitor your API usage and token costs.
- Visual Click Markers: Highlights the target of the AI's last action, making it easier to debug and correct if the model misses the mark.
Measured star growth
- in 24h
- 0
- in 7 days
- Collecting history
- in 15 days
- Collecting history
- in 1 month
- Collecting history
Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.
README.md
OriginalOperator - On-Device AI Agent for Android UI Automation
Call AI APIs directly from your phone. No PC connection required—let the AI operate your device untethered!
Video Demo (Android 17, Stock Android for PC)
https://github.com/user-attachments/assets/6dd29d4e-1f6b-48eb-8dc8-359d484581bc
Key Features
- Virtual Screen Support: The AI operates on a background virtual screen without occupying your main display. You can chat, browse, or watch videos while the AI works in the background.
- Accessibility-based Layout Parsing: Leverages Android Accessibility Services to parse screen layouts, providing coordinates of UI elements and richer context for the AI.
- Token Consumption Tracking: Built-in stats to monitor your API usage and token costs.
- Visual Click Markers: Highlights the target of the AI's last action, making it easier to debug and correct if the model misses the mark.
- Broad App Compatibility: Supports launching and controlling any user-installed application on the device.
- Status Bar Integration: Task execution status integrates into the system status bar / dynamic island on devices running Android 16+.
- Highly Optimized Performance: Deeply optimized virtual screen execution ensures smooth background operations.
Quick Start
1. Download and Install
Go to Releases and download the APK compatible with your device architecture:
- Android 8.0+ (standard devices):
app-arm64Minsdk26-release.apk - Android 10+:
app-arm64Minsdk29-release.apk - Android 15+:
app-arm64Minsdk35-release.apk - Emulators:
app-x86_64-release.apk - Unsure? Download:
app-universal-release.apk
2. Configuration
Follow the in-app prompts to configure your API settings. Fill in the URL, API Key, and the Model Name provided by your LLM/VLM service provider.
- Tip: Click the arrow icon next to the model name field. If supported by your model provider, the app can automatically search and fetch available models using the keyword you entered.
3. Send Commands
- Tap the input box at the bottom, enter your instructions, and tap the "arrow" icon to send.
- If you manually intervene and take over the device, press the "arrow" button again to let the AI resume execution.
Tested Devices
| Android Version | OS / ROM | Limitations / Status |
|---|---|---|
| 17 | Stock Android for PC | When the screen is locked and turned off, the screen goes black and touch input cannot be injected. |
| 16 | ColorOS 16 | Fully functional; no issues observed. |
| 15 | HyperOS 2 | Touch input injection fails when the screen is turned off. |
| 10 | HarmonyOS 3 | Cannot use virtual screens. |
Star History
Explore related projects
affaan-m/ECC
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
301,942 starsNousResearch/hermes-agent
The agent that grows with you
262,948 starsmultica-ai/andrej-karpathy-skills
A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.
232,510 starsfirecrawl/firecrawl
Supercharge your AI agents with data from the web and beyond. Building the library for superintelligence. 🔥
200,932 stars