NVIDIA BioNeMo Agent Toolkit: skills for life sciences
NVIDIA BioNeMo Agent Toolkit
NVIDIA skills for biology and chemistry: the agent predicts protein structures, docks molecules and designs proteins
- 504 ★ on the repository
- Last commit
Extended access
Why this level
- The skills let the agent run Bash, read and write files and send sequences to the NVIDIA hosted API
- The agent reads NGC_API_KEY from the environment and passes it in a request header
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/nvidia-bionemo-agent-toolkitDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skills let the agent run Bash, read and write files and send sequences to the NVIDIA hosted API
- The agent reads NGC_API_KEY from the environment and passes it in a request header
This is third-party code. Review the repository files before installing.
What it does
The NVIDIA repository collects skills that guide an agent through BioNeMo models and libraries. The NIM skills call NVIDIA microservices through a hosted API or a local Docker container. Boltz-2 and OpenFold3 predict structures of proteins and complexes, DiffDock computes docking poses for small molecules, GenMol and MolMIM generate molecules, Evo 2 works with DNA sequences, and ProteinMPNN and RFdiffusion design proteins. End-to-end workflows chain the steps. The drug-discovery-pipeline skill goes from molecule generation in GenMol through DiffDock docking to affinity scoring in Boltz-2, and msa-structure-prediction-pipeline builds a multiple sequence alignment first, then a structure in OpenFold3. Skills for open models, including RFD4-Proteina, Proteina-Complexa, KERMT and CodonFM, and skills for the Parabricks, nvMolKit and cuEquivariance libraries work with locally installed models and software. Each skill is a folder with a SKILL.md that sets out the order of work: pick a tool, prepare inputs, run it, check the result and explain it.
Who it is for. Researchers in bioinformatics and computational chemistry who run calculations through Claude Code or Codex and do not want to build NIM requests by hand.
Good fit when
- You have a protein sequence and need a structure in mmCIF format with a confidence score
- Ligand candidates must be generated, docked to a target and sorted by predicted affinity
- A de novo protein binder design project needs a route through RFdiffusion, ProteinMPNN or Proteina-Complexa
- Sequencing data is processed with Parabricks and the agent picks parameters for its commands
- ADMET properties for a set of molecules are predicted by the KERMT model
Not a fit when
- You need clinical or diagnostic interpretation: the skills run computational models and do not replace lab verification
- You have no NGC key for the hosted API and no prepared environment for a local run: without one of the two there is nothing to run the calculation on
- You need a verified result today: the OpenFold3 runs and the MSA-Search to OpenFold3 workflow are still pending validation
Example request
Use Boltz-2 to predict a protein-ligand complex from my sequence and SMILES and show the pIC50 affinity estimate.Limitations
The models themselves are not included in the repository. The NVIDIA hosted API requires an NGC_API_KEY, and skills for open models, such as RFD4-Proteina, need a separate model checkout and a GPU. Installing the skills does not create that environment.
Licenses differ: code is Apache-2.0, skill texts are CC-BY-4.0, and some bundles keep their own terms. Third-party data referenced in the skills has its own usage rules.
How to disable. Delete the folders of unneeded skills from where the installer put them, for example ~/.claude/skills/, or disable the plugin in Claude Code. Remove the NGC_API_KEY variable from the environment by hand.
FAQ
How do I install the BioNeMo skills in Claude Code?
Run npx skills add NVIDIA-BioNeMo/bionemo-agent-toolkit --skill boltz2-nim --agent claude-code, substituting the skill you need. Without the --skill flag the installer lets you pick skills and the target interactively. The Claude Code plugin can also be installed through the .claude-plugin/marketplace.json file.
Do I need an NVIDIA key?
For the hosted API, yes: NGC_API_KEY is sent in the Authorization header. The skill only checks that the variable is set and never prints its value. A local NIM container needs no key in requests once it is running, but the key is needed when the image and model are downloaded.
Can I work without the NVIDIA cloud?
Yes, many skills have a local Docker container mode at an address like http://localhost:8000. Skills for open models and libraries run calculations on your own machine.
Which skill fits ligand screening?
drug-discovery-pipeline: it runs molecule generation in GenMol, DiffDock docking and Boltz-2 affinity scoring as one workflow.
Which tools does the set cover?
Among NIM there are Boltz-2, DiffDock, Evo 2, GenMol, MolMIM, MSA-Search, OpenFold2 and OpenFold3, ProteinMPNN, RFD4-Proteina and RFdiffusion. Open models Proteina-Complexa, KERMT and CodonFM and the Parabricks, nvMolKit and cuEquivariance libraries come separately.
Discussion
No questions or reviews yet.
Related
SkillsA Google DeepMind skill collection for science tasks: access to biology and chemistry databases, literature search and structural biology
Around 200 skills for Claude Code on bioinformatics, chemistry and biostatistics, from RNA-seq analysis to molecular docking
NVIDIA's skill for Claude Code and Codex: the agent turns a dataset description into a config, validates it, previews it and generates synthetic data
166 skills for scientific work: bioinformatics, cheminformatics, clinical data, geospatial analysis and 100+ databases
A skill that turns SMILES into a Mermaid graph of named atoms, so the agent edits molecules without hand-writing SMILES
A Claude Code plugin and MCP server: over 1000 scientific tools for biology, chemistry and medicine, data-check commands and routing skills
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| NVIDIA BioNeMo Agent Toolkit: skills for life sciences (this card) | yes | 504 | |
| NVIDIA Agent Skills | no | 3,878 | NVIDIA's general skill catalog for CUDA, RAG and robotics, with no biology or chemistry |
| Life Sciences: Anthropic's marketplace for Claude Code | yes | 642 | an Anthropic marketplace with PubMed, ChEMBL and ClinicalTrials.gov search, without running computational models |
README in short
Skills are arranged by folder: nim-skills for NIM microservices, open-models-skills for open models, library-skills for libraries and workflows for end-to-end scenarios. Skill lift is measured with ACES and SkillEvaluator by running one request with and without the skill. The repository's CI has a separate skill security check.
README badge
Original SKILL.md
---
name: boltz2-nim
description: >
Use Boltz2 NIM for biomolecular structure prediction and binding affinity. Invoke for Boltz2, protein structures, protein-ligand/DNA/RNA complexes, SMILES or CCD ligands, pIC50/IC50 affinity scoring, mmCIF output, hosted NVIDIA API calls, or local Docker deployment.
license: Apache-2.0 AND CC-BY-4.0
compatibility: "requests>=2.28"
allowed-tools: Bash, Read, Write, AskUserQuestion
---
# Boltz2 NIM
Predict biomolecular structures and optional ligand affinity. Use this
guide for first-pass hosted/local usage; load supplemental files only when
needed:
- `references/api.md`: exact endpoints, schemas, Docker flags, response fields.
- `references/science.md`: purpose, strengths, limitations, and handoffs.
- `references/parameters.md`: prediction, sampling, MSA, template, affinity tuning.
- `references/validation.md`: mmCIF, confidence, affinity, and chemistry checks.
- `references/examples.md`: compact hosted/local payload patterns.
## Instructions
Read credentials from the environment only when needed. Check presence with
`bool(os.getenv("NGC_API_KEY"))`; keep key values and Authorization headers out of
terminal output, logs, saved artifacts, and the final response. Avoid environment
dumps when diagnosing authentication. If the hosted key is absent, report the
missing variable before submitting a request.
### Choose Mode
Ask only when context is unclear:
> Hosted NVIDIA API or local Docker NIM?
- Hosted: `https://health.api.nvidia.com/v1/biology/mit/boltz2/predict`
- Local: `http://localhost:8000/biology/mit/boltz2/predict`
Hosted requests use `Authorization: Bearer $NGC_API_KEY`. Supported local Docker
startup uses `NGC_API_KEY` (or `NVIDIA_API_KEY` via the preflight) for
registry login, entitlement checks, and first-run model downloads; pass it
into the container with `-e NGC_API_KEY`. Local inference requests use no
auth header after readiness. Warm-cache key-free startup varies by
image/version and should not be assumed.
## Local Docker
For local setup answers, copy the preflight below before `docker login`,
`docker run`, readiness, and the no-auth local request. Do not invent a cache
default or drop the `.env` load or `NVIDIA_API_KEY` fallback.