NVIDIA’s BioNeMo Agent Toolkit bridges AI agents and biomolecular research
AI agents are stepping into scientific research—but biology is not software engineering. A new open-source toolkit from NVIDIA aims to bridge that gap by turning biomolecular models into reliable, callable skills for AI agents in drug discovery.
NVIDIA’s BioNeMo Agent Toolkit packages core capabilities—protein folding, molecular docking, generative chemistry, genomics, protein design, and biomarker discovery—as documented agent tools. The platform combines accelerated inference services through NVIDIA NIM microservices and BioNeMo open models, supported by libraries like cuEquivariance for structure modeling and Parabricks for genomics. Each skill is packaged with clear documentation of purpose, inputs, parameters, expected outputs, and failure modes, making it easier for agents to discover, select, and invoke the right tool for the task.
From models to agent-ready skills
The toolkit organizes skills into three groups: nim-skills for hosted NIM endpoints, open-models-skills for BioNeMo open models, and library-skills for utility functions. A workflows folder includes multi-step meta-skills that chain tools—such as a generative protein binder design pipeline combining RFdiffusion, ProteinMPNN, and OpenFold3. Each skill is defined by a SKILL.md file containing YAML frontmatter, instructions, references, and optional scripts, acting as machine-readable documentation the agent can interpret.
Flexibility for different workflows
NVIDIA highlights two deployment options: hosted NIM endpoints for quick access and local NIM deployments for iterative experimentation. Agents can call models like Boltz-2, DiffDock, GenMol, ProteinMPNN, MSA Search, RFdiffusion, and Evo 2 using consistent prompt patterns. Installation is streamlined via the open-source skills CLI, allowing agents or users to browse and add skills interactively or install specific ones—such as boltz2-n—for targeted tasks.
Source: MarkTechPost. AI-assisted editorial synthesis — TechnoExpress.