BioNeMo Agent Toolkit v0.1.0 — Initial Release

Turn any agent into a life-science expert with NVIDIA BioNeMo skills — protein folding, molecular docking, generative chemistry, genomics, protein design, and biomarker discovery, packaged as ready-to-call agent skills with instructions, scripts, and references.

Workflow skills

generative_protein_binder_design — end-to-end de novo protein binder design against a target, in two routes:

• protein-binder-design — composes BioNeMo NIMs: RFdiffusion (backbones) → ProteinMPNN (sequences) → Boltz-2 / OpenFold3 (co-fold + score), with optional MSA-Search for higher-quality folding.
• complexa-binder-design — uses Proteina-Complexa to co-design binder sequence + structure, then independently validates and ranks each binder by refolding the complex with Boltz-2 / OpenFold3 (interface confidence, pLDDT, stability, hotspot contact).

NIM skills

BioNeMo NIM microservices — each runs against the hosted API or a self-hosted local container.

Skill	What it does	NIM
boltz2-nim	Biomolecular structure + binding affinity	mit/boltz2
diffdock-nim	Small-molecule docking / pose prediction	mit/diffdock
evo2-nim	DNA generation & variant scoring (Evo 2)	arc/evo2-40b
genmol-nim	De novo molecule generation, scaffold decoration	nvidia/genmol-generate
molmim-nim	Latent-space molecule generation/optimization	nvidia/molmim-generate
msa-search-nim	Multiple sequence alignments (ColabFold)	colabfold/msa-search
openfold2-nim	Monomer protein structure prediction	openfold/openfold2
openfold3-nim	Biomolecular complex structure prediction	openfold/openfold3
proteinmpnn-nim	Inverse folding / sequence design	ipd/proteinmpnn
rfdiffusion-nim	De novo backbone & binder design	ipd/rfdiffusion

Meta-NIM skills

Multi-NIM compositions:

Skill	What it does
drug-discovery-pipeline	GenMol → DiffDock → Boltz-2 (generate molecules → dock to target → score affinity)
msa-structure-prediction-pipeline	MSA-Search (ColabFold) → OpenFold3 (align → predict structure)

Open-model skills

Skills wrapping open BioNeMo models you run yourself.

Model	What it does	Component Source
Proteina-Complexa	Co-design of protein binder sequence + structure with reward-guided test-time search	NVIDIA-Digital-Bio/Proteina-Complexa
KERMT	GROVER/cMIM molecular encoder for embeddings, property prediction & ADMET	NVIDIA-BioNeMo/KERMT

Library skills

Atomic skills wrapping NVIDIA accelerated libraries.

Skill	What it does	Component Source
parabricks	pbrun GPU genomics command guidance (FASTQ/BAM, variant calling, RNA-seq)	clara-parabricks-workflows
genomics-workflow-acceleration	Plan GPU-acceleration of existing genomics pipelines with Parabricks	clara-parabricks-workflows
nvMolKit	GPU-accelerated, RDKit-style cheminformatics (fingerprints, similarity, conformers)	NVIDIA-BioNeMo/nvMolKit
cuEquivariance	Build equivariant NN primitives (segmented tensor products, CG coefficients)	nvidia/cuequivariance