NVIDIA BioNeMo Agent Toolkit Turns Biomolecular Models Into Callable Skills for AI Agents in Drug Discovery

for AI Agents in Drug Discovery">

AI scientists are becoming a new interface for scientific computing. These agents read papers, write code, generate hypotheses, call APIs, and inspect files. But science is not software engineering. No test suite turns green when a hypothesis is correct. Discovery stays iterative, uncertain, and grounded in the physical world.

That gap is what NVIDIA is targeting. NVIDIA published a hands-on walkthrough for its BioNeMo Agent Toolkit . The argument is direct. A general coding agent pointed at biology will not produce new medicines. In biomolecular research, an agent’s ceiling is set by the tools it can use reliably, correctly, and efficiently.

The BioNeMo Agent Toolkit is an open-source repository of ‘skills’ for AI agents. Each skill turns an NVIDIA biomolecular model into a tool an agent can call. The toolkit packages protein folding, molecular docking, generative chemistry, genomics analysis, protein design, and biomarker discovery.

NVIDIA frames the platform in two parts. The first is an accelerated tool layer. NVIDIA NIM (NVIDIA Inference Microservices) and BioNeMo open models deliver core capabilities as callable services. These are accelerated by libraries such as cuEquivariance for structure models and Parabricks for genomics. The second part is agent-ready interfaces. BioNeMo Skills package each capability so an agent can use it.

A skill documents the model’s purpose, required inputs, optional parameters, expected artifacts, and failure modes. Model Context Protocol (MCP) server wrappers expose open models not yet packaged as NIM. Together, this lets an agent discover, select, invoke, and interpret biomolecular models on its own.

The repository groups skills into nim-skills , open-models-skills , and library-skills . A workflows folder holds multi-step meta-skills. One example is generative_protein_binder_design , which chains RFdiffusion → ProteinMPNN → OpenFold3.

Every skill is a directory with a SKILL.md file. It holds YAML frontmatter plus instructions, optional references, and optional scripts. An agent reads it like documentation, then acts on it.

The prompt pattern stays the same across models. The NVIDIA’s post uses OpenFold3. The same shape applies to other NIMs for biology. These include Boltz-2, DiffDock, GenMol, ProteinMPNN, MSA Search, RFdiffusion, and Evo 2. You name the skill, the input, and the endpoint.

Installation pulls skills through the open-source skills CLI: