AI

Glint

A screenshot-to-AI assistant for Linux (Wayland) — capture a screen region, ask AI about it, and get real-time responses in a floating window.

Source Code
#ai #linux #wayland #python #screenshot

What is Glint?

Glint is a screenshot-to-AI assistant for Linux, designed for Wayland compositors like Hyprland and Sway. Think of it like Google’s Circle to Search — press a hotkey, select an area of your screen, and ask an AI about what you see. It’s fast and lightweight. Currently it only uses cloud models but I plan to add local model support. It’s just hard to find good local models that support images that are fast, and who wants to wait for processing. APIs are fairly cheap for this, all of my testing ran me about $0.05 in fees.

Glint Screenshot

How It Works

  1. Press a hotkey (e.g. Super+Shift+S)
  2. Select a screen region using slurp
  3. A floating window appears with a preview of your screenshot
  4. Type a question about what you captured
  5. Get a real-time AI response streamed directly into the window
  6. Ask follow-ups — the session stays open for continued conversation

Supported AI Providers

Glint currently supports two AI backends:

  • xAI — uses the grok-4.20-0309-reasoning model by default
  • OpenRouter — Grok by default, but can choose any

Local AI support is planned for a future release, so you’ll be able to run everything entirely on your own hardware.

Tech Stack

Glint is built with Python and GTK4/libadwaita for a native Linux look and feel. It uses grim and slurp for Wayland-native screen capture, and httpx for streaming AI responses. Configuration lives in ~/.config/glint/config.toml.

Installation

git clone https://github.com/Cube-Oakley/glint.git
cd glint
bash install.sh

The installer checks dependencies, prompts for your AI provider and API key, and sets everything up. Then just bind the glint command to a hotkey in your compositor config.

← Back to all projects