showtell lets you point at something on your screen, say what you mean, and hand both to an AI agent. Press the shortcut once, frame the area, talk, press it again. It copies a small prompt with an image path and your words underneath.
For setup, see INSTALLATION.md.
It is meant for the moments where typing is worse than pointing:
- “this button here”
- “why does this spacing feel wrong?”
- “fix the thing inside this part of the screen”
- “explain what is happening in this graph”
- “make this less ugly”
The clipboard output is plain text:
Image:
/path/to/screenshot.png
Instructions:
Make this panel less cramped.No special paste target. No rich clipboard tricks. The image is saved on disk, and the agent gets the path.
Press Super+Ctrl+D to start. A frame appears around the part of the screen you are pointing at. Scroll while holding Super+Ctrl to make the frame bigger or smaller. Speak. Press Super+Ctrl+D again.
When the transcript is ready, showtell says Ready and leaves the prompt in your clipboard.
This is a Linux/Hyprland tool. Other desktops may work later; Hyprland is the path for now.
It uses:
grimfor screenshotspw-recordfor audiowl-copyfor the clipboard- Codex transcription by default
- a compatible multipart transcription endpoint if you want to bring your own
MIT.