zebinlu7-a11y/screen-flow-ai-agent

This is a Python desktop app that keeps a small floating chat panel visible on top of all your other windows. When you run it, the panel sits in the corner of your screen and you can type questions into the bottom bar at any time, getting AI answers that stream in word by word. The main feature is screenshot-based questioning. Press Ctrl+D to draw a selection box around anything on your screen and the app captures that region as an image. You can grab multiple regions in one session, then submit them all together with a typed question. The AI reads the captured images and responds based on what it sees, whether that is code, a chart, a UI design, or any other visual content. There is also a separate OCR mode triggered with Ctrl+R. Instead of sending the screenshot to the AI, this route runs it through Tencent Cloud's text-recognition service and extracts the readable text. The result lands in the chat input so you can copy it, edit it, or ask a follow-up question about it. The Tencent OCR service comes with a free allowance of 1,000 calls per month. The floating window itself is adjustable: resize it, drag it anywhere on screen, make it semi-transparent, or hide it entirely with Ctrl+F. A system tray icon keeps the app accessible without cluttering your taskbar. Three model tiers (mini, lite, pro) can be switched at any time from inside the window. API keys and proxy settings are entered through a settings button in the UI, so no manual config file editing is needed. Under the hood the project uses Python with LangGraph managing the conversation flow. The AI backend is the DouBao multimodal model from ByteDance's VolcEngine platform, which handles both text and images in the same request. A pre-built Windows executable is available in the releases section if you prefer not to run from source.