👁️ OpenEyes-Live
一个模块化、可插拔、端云协同的端侧视频AI基础设施。 —— 把最新的端侧多模态模型，变成可独立下载、按需加载、自由组合的AI引擎。
---
🎬 See It In Action
A 5-second demo is worth a thousand lines of code.
| 👇 Real-time Scene Understanding | 🔔 Proactive Alert | | :---: | :---: | | | | | Camera sees a cat → AI says "An orange cat is lying on the sofa" | Camera sees boiling water → AI alerts "Water is boiling!" |
[!IMPORTANT] Upload your actual demo GIF here before sharing! 1. Record a 5-second video using OpenEyes-Live 2. Convert to GIF using ffmpeg -i input.mp4 -vf "fps=15,scale=640:-1" -loop 0 output.gif 3. Upload to your GitHub repo's /assets/ folder or imgur 4. Replace the image links above
No demo yet? Run it now and record one — a rough demo beats a perfect README any day.
---
📚 Full Documentation
| Document | Description | Audience | | :--- | :--- | :--- | | 📖 PROJECTDOC.md | Vision, scenarios, roadmap, user personas | All users | | 🏗️ ARCHITECTURE.md | Scheduler, engines, MCP gateway, lifecycle | Developers | | 🔪 VIDEOPIPELINE.md | Video engine pipeline: sampler/filter/encoder/compressor | Developers | | 🎤 AUDIOPIPELINE.md | Audio engine pipeline: VAD/ASR/speaker | Developers | | 📊 PERFORMANCEBENCHMARK.md | Device compatibility, benchmark data | Developers, decision-makers | | 📱 QUICKSTARTMOBILE.md | IP Webcam + Termux setup guide | All users | | 🔧 APIREFERENCE.md | Complete Python API reference | Developers | | 📐 ENGINESPEC.md | Engine interface specification for contributors | Contributors | | 🤝 CONTRIBUTING.md | Contribution guidelines | Contributors | | 📜 CODEOFCONDUCT.md | Community standards | All community members | | 📋 CHANGELOG.md | Release notes and version history | All users |
---
🤔 Why OpenEyes-Live?
Got old phones sitting in a drawer?
- Screen still works, camera still works, WiFi still works
- But the OS is too old, apps won't install — just collecting dust
What if I told you that with just 80MB of storage, that "e-waste" could become an AI device that sees, hears, and understands the world?
OpenEyes-Live's promise is simple:
| Promise | Description | | :--- | :--- | | 📦 80MB+ | Smallest sub-engine is just 80MB — fits on almost any phone | | 🔪 Video Pipeline | Sampler → Filter → Encoder → Compressor — each pluggable | | 🎤 Audio Pipeline | VAD + ASR + Speaker Recognition — let AI "hear and identify" | | 🧩 On-Demand | Download only the engines you need. No wasted storage. | | 🔒 Fully Local | Camera feed never leaves your device. Zero uploads. | | 🌏 Global Mirror | Models hosted on Hugging Face + ModelScope for fast downloads in China | | 💰 100% Free | No API keys, no subscriptions, no hidden costs |