Skip to content

Installation

Version 0.1.0 includes GPU image recognition and COSMIC capture. Follow the source installation below to use the workspace's CUDA-enabled PyTorch setup. Version 0.0.1 provides configuration commands only.

Requirements

  • Linux with an NVIDIA GPU and a working NVIDIA driver.
  • Python 3.14 or later and uv.
  • mise for the repository's development tools.
  • COSMIC's cosmic-screenshot for interactive capture.
  • wl-clipboard on Wayland, or xclip/xsel on X11, for clipboard output.
  • notify-send, usually supplied by libnotify, for desktop notifications.

Verify that nvidia-smi reports your GPU before starting the model service. GPU memory requirements depend on the selected model and image size.

Install from source

git clone https://github.com/HYP3R00T/formulens.git
cd formulens
mise trust
mise install
uv sync --package formulens --extra ocr
mise run install-cli

The workspace selects CUDA-enabled PyTorch from the official CUDA 13.0 wheel index. The ocr extra installs model dependencies. The CLI is installed in editable mode, so source changes are available to new CLI processes. After dependency changes, rerun mise run install-cli to refresh the separate CLI environment; syncing the workspace alone does not update it.

If the CLI is not found, run uv tool update-shell and open a new terminal.

Prepare the model

formulens config init
formulens model-download

model-download downloads and loads the selected model to verify that it works. It requires an operational CUDA GPU and exits after verification. You can skip this step: the daemon also downloads missing model files during its first start.

By default, weights are cached in ~/.config/formulens/models. Subsequent starts reuse those files. See configuration to change the location or model, then follow the usage guide.