back to projects

Dip — Give LLMs Eyes

A tool that gives LLMs real-time visual context of your screen, so you never have to describe what's on it.

PythonGoogle Gemini APIScreen Capture

Dip streams real-time screen context into an LLM: it captures what's on your screen and feeds it to the model, so describing on-screen content by hand becomes unnecessary. A lightweight client captures the screen and a Python backend pipes it into Gemini for pair-programming and debugging workflows.

  • 1st Place, Google DeepMind Gemini Track
  • 400+ participants at the Cursor AI Hackathon Hamburg
  • Real-time screen capture as live LLM context

Learned how to feed multimodal models with real-time image data while keeping latency low, working with screen-capture APIs and the Gemini Vision API.