Vision System
Real-time screen capture, frame memory, and scene analysis.
The Vision System gives ZYRAXON continuous visual awareness of your machine. It lives in packages/zyraxon/src/screen/ and powers both Auto Screen Vision and full Vision Mode.
Architecture
| Module | Responsibility |
|---|---|
| capture-engine.ts | Cross-platform screenshot capture |
| frame-memory.ts | Frame storage and memory management |
| frame-processor.ts | Frame processing pipeline |
| vision-analyzer.ts | Scene analysis and understanding |
| vision-context.ts | VisionContextManager singleton |
| vision-mode.ts | VisionMode orchestrator |
| auto-vision.ts | Automatic vision injection into every prompt |
How it works
- 01You send a message.
- 02The session layer checks whether the active agent uses vision.
- 03A screenshot is captured and converted to base64.
- 04The image is injected as content into the AI message.
- 05The model receives and processes the screenshot.
- 06The agent replies with full screen understanding.
Vision Mode capabilities
- ▸24/7 continuous screen streaming
- ▸Frame memory with scene-change detection
- ▸Activity tracking and intelligent recall
- ▸Timeline recall -- ask what happened on screen minutes ago
★
Auto Vision
In modes that enable it, vision is injected automatically into every prompt, so the agent sees what you see without being asked.