MoleculeDesk documentation

MoleculeDesk is an open-source package manager and runtime for running molecular AI models locally. The CLI provides a consistent interface for supported models, while model pages distinguish working beta adapters from planned integrations.

curl -fsSL https://moleculedesk.com/install.sh | sh
 
moldesk doctor
moldesk install proteinmpnn
moldesk list --installed

MoleculeDesk is currently in early development. System detection, model discovery, and installing/running Boltz-2, ProteinMPNN, and LigandMPNN already work end-to-end; DiffDock-L is coming soon (its adapter is implemented but not yet live-verified), and more model adapters (AlphaFold, ColabFold, Chai) are in progress. Every page below states a model's current status plainly.

Where to start

• Run Boltz-2 locally — install and run a real example.

• Molecular AI on Apple Silicon — verified platform support and current limits.

• Installation — install the CLI and confirm your machine is ready.

• CLI usage — the day-to-day doctor / install / run / uninstall workflow.

• Architecture — how the CLI, core engine, runtime, and model registry fit together.

• Model compatibility & registry — how MoleculeDesk decides whether a model can run on your machine.

Models

Each supported (or planned) model has its own page with setup details, requirements, and use cases:

• Boltz-2 — joint structure & binding-affinity prediction (beta)
• DiffDock-L — blind small-molecule docking (coming soon)
• BindCraft2 — bounded local protein-binder campaigns (planned; Linux/NVIDIA)
• OpenDDE Preview — all-atom biomolecular co-folding (beta, upstream preview)
• ProteinMPNN — structure-conditioned sequence design (beta)
• LigandMPNN — ligand- and multimer-aware sequence design (beta)
• AlphaFold — structure prediction (coming soon)
• ColabFold — fast structure prediction (coming soon)
• Chai — biomolecular structure prediction (coming soon)