NVIDIA BioNeMo NIMs (hosted)
The ten BioNeMo NIM adapters run against NVIDIA's hosted endpoints with the user's own
NVIDIA API key. Every request goes through the scientific_capability tool; nothing here
installs software or starts compute. The capability ids are boltz2, openfold2,
openfold3, msa-search, diffdock, proteinmpnn, rfdiffusion, genmol, molmim
and evo2.
Before the first request
scientific_capability with action: "list" shows the catalog and each entry's maturity.
action: "doctor", id: "<id>" says whether the NVIDIA key is connected and the endpoint
is reachable. Without a key, stop and tell the user: connect one under
Customize → Connectors → NVIDIA API (an nvapi-… key from build.nvidia.com); requests
are billed to that NVIDIA account under NVIDIA's terms. Do not look for the key in the
environment, in compute targets, or in Modal secrets; the hosted route is separate from the
self-hosted BioNeMo containers that protein-binder-design runs on Modal.
action: "describe", id: "<id>" returns the hosted request_schema, the endpoint, the API
schema version and the terms. Build payload from that schema exactly; field names and
enumerations differ between NIMs (a sequence list for Boltz-2, a PDB string plus ligand
SMILES for DiffDock, an alignment for OpenFold2).
Each request
action: "plan", id, payload validates every cross-field requirement locally. Nothing is
sent and nothing is spent. Fix what it reports before going on.
action: "start", id, payload, name, purpose sends the request. The user sees an approval
card bound to this exact request: endpoint, schema version, payload size, request hash and
what leaves the device (sequences, structures, ligands). The approval is one-time; a
changed payload asks again. Say in purpose what the result is for.
action: "wait", job_id (or status) until it settles; long predictions poll NVIDIA's
status endpoint for you. action: "artifacts", job_id lists the delivered files with
their SHA-256; read a structure or JSON from there, and save what the user keeps with
artifact save_file.
- Report the model, the API schema version from
describe, the request hash and the
artifact hashes with the result. A predicted structure or pose is a prediction: give its
confidence (pLDDT, iptm, DiffDock confidence) and never call it experimental validation.
Choosing a NIM
| Need |
Capability |
| Protein or complex structure from sequence, ligands and ions allowed |
boltz2 |
Monomer structure from an MSA (msa-search first) |
openfold2 |
| Complex structure with templates |
openfold3 |
| Protein–ligand poses for a given pocket |
diffdock |
| Sequences for a fixed backbone |
proteinmpnn |
| New backbones or binders against a target |
rfdiffusion |
| Small molecules from a fragment or property target |
genmol, molmim |
| DNA sequence generation and scoring |
evo2 |
For a full binder campaign (RFdiffusion → ProteinMPNN → Boltz-2 refold → ranking), load
protein-binder-design, which chains these the way the BioNeMo Agent Toolkit does.
1---2name: bionemo-nims3description: Run NVIDIA BioNeMo NIMs through the hosted scientific_capability tool with your own NVIDIA API key. Boltz-2 and OpenFold2/OpenFold3 structure prediction, MSA Search alignments, DiffDock docking, ProteinMPNN sequence design, RFdiffusion backbones, GenMol and MolMIM molecule generation, Evo 2 genomic sequence modeling. Use when a request names one of these models, "BioNeMo", "NIM", or asks for a hosted structure, docking or design prediction; not for self-hosted containers on Modal (see protein-binder-design).4summary: "Hosted NVIDIA BioNeMo NIMs via scientific_capability: list, describe, plan, start, wait, artifacts; one approval per request."5category: biology6tags:7 - bionemo8 - nvidia9 - structure-prediction10 - docking11 - protein-design12role: workflow13allowed-tools: [scientific_capability, read, artifact]14license: CC-BY-4.015metadata:16 source: https://skillmd.com/skills/synthetic-sciences/bionemo-nims17 skill-author: Synthetic Sciences18 upstream: NVIDIA-BioNeMo/bionemo-agent-toolkit19 upstream-license: CC-BY-4.0 (skills) / Apache-2.0 (code)20version: 1.0.021---2223# NVIDIA BioNeMo NIMs (hosted)2425The ten BioNeMo NIM adapters run against NVIDIA's hosted endpoints with the user's own26NVIDIA API key. Every request goes through the `scientific_capability` tool; nothing here27installs software or starts compute. The capability ids are `boltz2`, `openfold2`,28`openfold3`, `msa-search`, `diffdock`, `proteinmpnn`, `rfdiffusion`, `genmol`, `molmim`29and `evo2`.3031## Before the first request3233- `scientific_capability` with `action: "list"` shows the catalog and each entry's maturity.34 `action: "doctor", id: "<id>"` says whether the NVIDIA key is connected and the endpoint35 is reachable. Without a key, stop and tell the user: connect one under36 **Customize → Connectors → NVIDIA API** (an `nvapi-…` key from build.nvidia.com); requests37 are billed to that NVIDIA account under NVIDIA's terms. Do not look for the key in the38 environment, in compute targets, or in Modal secrets; the hosted route is separate from the39 self-hosted BioNeMo containers that `protein-binder-design` runs on Modal.40- `action: "describe", id: "<id>"` returns the hosted `request_schema`, the endpoint, the API41 schema version and the terms. Build `payload` from that schema exactly; field names and42 enumerations differ between NIMs (a sequence list for Boltz-2, a PDB string plus ligand43 SMILES for DiffDock, an alignment for OpenFold2).4445## Each request46471. `action: "plan", id, payload` validates every cross-field requirement locally. Nothing is48 sent and nothing is spent. Fix what it reports before going on.492. `action: "start", id, payload, name, purpose` sends the request. The user sees an approval50 card bound to this exact request: endpoint, schema version, payload size, request hash and51 what leaves the device (sequences, structures, ligands). The approval is one-time; a52 changed payload asks again. Say in `purpose` what the result is for.533. `action: "wait", job_id` (or `status`) until it settles; long predictions poll NVIDIA's54 status endpoint for you. `action: "artifacts", job_id` lists the delivered files with55 their SHA-256; `read` a structure or JSON from there, and save what the user keeps with56 `artifact save_file`.574. Report the model, the API schema version from `describe`, the request hash and the58 artifact hashes with the result. A predicted structure or pose is a prediction: give its59 confidence (pLDDT, iptm, DiffDock confidence) and never call it experimental validation.6061## Choosing a NIM6263| Need | Capability |64| --- | --- |65| Protein or complex structure from sequence, ligands and ions allowed | `boltz2` |66| Monomer structure from an MSA (`msa-search` first) | `openfold2` |67| Complex structure with templates | `openfold3` |68| Protein–ligand poses for a given pocket | `diffdock` |69| Sequences for a fixed backbone | `proteinmpnn` |70| New backbones or binders against a target | `rfdiffusion` |71| Small molecules from a fragment or property target | `genmol`, `molmim` |72| DNA sequence generation and scoring | `evo2` |7374For a full binder campaign (RFdiffusion → ProteinMPNN → Boltz-2 refold → ranking), load75`protein-binder-design`, which chains these the way the BioNeMo Agent Toolkit does.