CLI usage

The typical MoleculeDesk workflow is: check your machine, install a model, confirm it's installed, then manage it.

moldesk doctor              # detect hardware, Python, Docker, CUDA
moldesk install proteinmpnn # install a model
moldesk list --installed    # confirm what's installed
moldesk uninstall proteinmpnn --yes

`moldesk doctor`

Detects your system and hardware, and explains compatibility for Python, Docker, CUDA, GPU, RAM, and disk against the models in the registry.

`moldesk list`

Shows every model in the registry, its status (available, beta, or planned), and — with --installed — which ones are currently installed on your machine.

moldesk list
moldesk list --installed

`moldesk install <model>`

Installs a model. By default this shows a plan and asks for confirmation; pass --yes to skip the prompt (useful in scripts/CI).

moldesk install proteinmpnn
moldesk install ligandmpnn --yes
moldesk install proteinmpnn --reinstall --yes

Under the hood, install gives Python models a pinned source checkout and an isolated uv-managed virtual environment, downloads any required checkpoints into a content-addressed cache (verified by SHA-256), and promotes the installation atomically — so a failed reinstall never leaves you with a broken model.

`moldesk uninstall <model>`

moldesk uninstall proteinmpnn --yes

Removes the model's installed environment and state, but retains shared cached downloads (like checkpoints) so another model — or a later reinstall — can reuse them.

`moldesk run <model> <input>`

Runs an installed model against an input file and reports the outcome.

moldesk run proteinmpnn structure.pdb
moldesk run proteinmpnn structure.pdb --output ./results  # also copy outputs here
moldesk run proteinmpnn structure.pdb --param <key>=<value>
moldesk run proteinmpnn structure.pdb --json               # machine-readable result

MoleculeDesk prints the run status and the path to each declared output as the run completes, plus the run directory it used. Pass --output <dir> to also copy the output files into a directory of your choice.

What's next

See each model's page for a runnable, model-specific example.