Getting started
Install & first run
Requirements: Python 3.10+, a GPU with 4 GB+ VRAM recommended (CPU works too), and ~5 GB disk for the model.
Install
pip install screenmind
Run
screenmind
Then open http://127.0.0.1:7777.
What happens on first run
- Prompts to install the AI packages (~2.5 GB one-time).
- Auto-detects your GPU and downloads the right
llama-serverbuild (CUDA/CPU) if it isn't found. - Opens the Model Hub — pick a Gemma 4 model with live progress right in the UI.
- Chat and Summary stay locked (🧠💤) until the model is ready, then auto-unlock.
- Creates
~/.screenmind/for all your data.
Developer install (from source)
Working on ScreenMind itself? Install it editable with the dev extras.
git clone https://github.com/ayushh0110/ScreenMind.git
cd ScreenMind
python -m venv venv
venv\Scripts\activate # Windows
# source venv/bin/activate # macOS/Linux
pip install -e ".[ai,dev]"
cd ScreenMind
python -m venv venv
venv\Scripts\activate # Windows
# source venv/bin/activate # macOS/Linux
pip install -e ".[ai,dev]"
Then set up a model with python -m screenmind.setup_llama (or use the Model Hub) and run the tests with pytest.
Configuration
Everything is configurable from the Settings tab (persisted to settings.json) or via a .env file — capture interval, analysis mode, performance/GPU-layer mode, blocked apps, retention, encryption, hotkeys, and integrations.
Tip: first run opens the Model Hub and fetches a Gemma 4 model for you — no terminal, no manual GGUF hunting. On a 4 GB GPU, pick Fast mode.
↯ back to the brain