Nvidia Nim Tools

Configuration File: nvidia_nim_tools.json Tool Type: Local Tools Count: 16

This page contains all tools defined in the nvidia_nim_tools.json configuration file.

Available Tools

NvidiaNIM_alphafold2 (Type: NvidiaNIMTool)

Predict protein 3D structure from amino acid sequence using DeepMind’s AlphaFold2 via NVIDIA NIM….

NvidiaNIM_alphafold2 tool specification

Tool Information:

• Name: NvidiaNIM_alphafold2

• Type: NvidiaNIMTool

• Description: Predict protein 3D structure from amino acid sequence using DeepMind’s AlphaFold2 via NVIDIA NIM. Returns PDB format structure. Requires NVIDIA_API_KEY. Input: sequence (amino acids), algorithm (jackhmmer/mmseqs2), databases (uniref90, mgnify, small_bfd). Long-running async operation.

Parameters:

• sequence (string) (required) Amino acid sequence to predict structure for (single letter codes)

• algorithm (string) (optional) MSA search algorithm. mmseqs2 is faster, jackhmmer is more sensitive

• databases (array) (optional) Sequence databases for MSA search: uniref90, mgnify, small_bfd

• e_value (number) (optional) E-value threshold for MSA search

• iterations (integer) (optional) Number of search iterations

• relax_prediction (boolean) (optional) Whether to perform structure relaxation

• skip_template_search (boolean) (optional) Skip template-based prediction

Example Usage:

query = {
    "name": "NvidiaNIM_alphafold2",
    "arguments": {
        "sequence": "example_value"
    }
}
result = tu.run(query)

NvidiaNIM_alphafold2_multimer (Type: NvidiaNIMTool)

Predict multi-chain protein complex structure using AlphaFold2-Multimer via NVIDIA NIM. Input: se…

NvidiaNIM_alphafold2_multimer tool specification

Tool Information:

• Name: NvidiaNIM_alphafold2_multimer

• Type: NvidiaNIMTool

• Description: Predict multi-chain protein complex structure using AlphaFold2-Multimer via NVIDIA NIM. Input: sequences (array of 1-6 protein sequences for complex). Returns predicted complex structure. Long-running async operation.

Parameters:

• sequences (array) (required) Array of 1-6 protein sequences (amino acid single letter codes) to predict as a complex

• databases (array) (optional) Sequence databases for MSA: uniref90, small_bfd

• relax_prediction (boolean) (optional) Whether to relax the predicted structure

Example Usage:

query = {
    "name": "NvidiaNIM_alphafold2_multimer",
    "arguments": {
        "sequences": ["item1", "item2"]
    }
}
result = tu.run(query)

NvidiaNIM_boltz2 (Type: NvidiaNIMTool)

Predict multi-molecular complex structure (proteins, DNA, RNA, ligands) using Boltz2 via NVIDIA N…

NvidiaNIM_boltz2 tool specification

Tool Information:

• Name: NvidiaNIM_boltz2

• Type: NvidiaNIMTool

• Description: Predict multi-molecular complex structure (proteins, DNA, RNA, ligands) using Boltz2 via NVIDIA NIM. Supports protein-protein, protein-DNA/RNA, protein-ligand complexes. Input: polymers (proteins/DNA/RNA), ligands (SMILES or CCD codes). Long-running async operation.

Parameters:

• polymers (array) (required) List of polymer molecules (1-12)

• ligands (array) (optional) Ligands specified by CCD code OR SMILES (max 20)

• recycling_steps (integer) (optional) Number of recycling steps (1-6)

• sampling_steps (integer) (optional) Number of sampling steps (10-1000)

• diffusion_samples (integer) (optional) Number of structure samples (1-5)

• output_format (string) (optional) No description

Example Usage:

query = {
    "name": "NvidiaNIM_boltz2",
    "arguments": {
        "polymers": ["item1", "item2"]
    }
}
result = tu.run(query)

NvidiaNIM_diffdock (Type: NvidiaNIMTool)

Blind molecular docking using DiffDock via NVIDIA NIM. Predict ligand binding poses without speci…

NvidiaNIM_diffdock tool specification

Tool Information:

• Name: NvidiaNIM_diffdock

• Type: NvidiaNIMTool

• Description: Blind molecular docking using DiffDock via NVIDIA NIM. Predict ligand binding poses without specifying binding pocket. Input: protein (PDB content or asset ID), ligand (SDF content or asset ID). For large files, use is_staged=true with asset IDs. Returns docking poses with confidence scores.

Parameters:

• protein (string) (required) Protein structure: PDB content directly OR asset ID (if is_staged=true)

• ligand (string) (required) Ligand structure: SDF/MOL2 content directly OR asset ID (if is_staged=true)

• ligand_file_type (string) (optional) Ligand file format

• num_poses (integer) (optional) Number of docking poses to generate (1-40)

• time_divisions (integer) (optional) Time divisions for diffusion

• steps (integer) (optional) Number of diffusion steps

• save_trajectory (boolean) (optional) Save diffusion trajectory

• is_staged (boolean) (optional) If true, protein and ligand are asset IDs (uploaded via NVCF assets API). If false, they are raw file content.

Example Usage:

query = {
    "name": "NvidiaNIM_diffdock",
    "arguments": {
        "protein": "example_value",
        "ligand": "example_value"
    }
}
result = tu.run(query)

NvidiaNIM_esm2_650m (Type: NvidiaNIMTool)

Generate protein sequence embeddings using ESM2-650M via NVIDIA NIM. 650 million parameter langua…

NvidiaNIM_esm2_650m tool specification

Tool Information:

• Name: NvidiaNIM_esm2_650m

• Type: NvidiaNIMTool

• Description: Generate protein sequence embeddings using ESM2-650M via NVIDIA NIM. 650 million parameter language model for protein sequences. Input: sequences (array of protein sequences, max 1024 AA each), format (npz/json). Output: embedding vectors in specified format. Returns binary npz file or JSON. Fast synchronous response.

Parameters:

• sequences (array) (required) Array of protein sequences (max 1024 amino acids each, valid: ARNDCQEGHILKMFPSTWYVXBOU)

• format (string) (required) Output format: npz (binary NumPy archive) or json

Example Usage:

query = {
    "name": "NvidiaNIM_esm2_650m",
    "arguments": {
        "sequences": ["item1", "item2"],
        "format": "example_value"
    }
}
result = tu.run(query)

NvidiaNIM_esmfold (Type: NvidiaNIMTool)

Fast alignment-free protein structure prediction using ESMFold via NVIDIA NIM. No MSA required - …

NvidiaNIM_esmfold tool specification

Tool Information:

• Name: NvidiaNIM_esmfold

• Type: NvidiaNIMTool

• Description: Fast alignment-free protein structure prediction using ESMFold via NVIDIA NIM. No MSA required - predicts directly from sequence. Input: sequence (max 1024 amino acids). Returns PDB structure. Fast synchronous response.

Parameters:

• sequence (string) (required) Amino acid sequence (max 1024 chars, valid: ARNDCQEGHILKMFPSTWYVXBOU)

Example Usage:

query = {
    "name": "NvidiaNIM_esmfold",
    "arguments": {
        "sequence": "example_value"
    }
}
result = tu.run(query)

NvidiaNIM_evo2 (Type: NvidiaNIMTool)

Generate DNA sequences using Evo2-40B via NVIDIA NIM. 40 billion parameter model trained on 9 tri…

NvidiaNIM_evo2 tool specification

Tool Information:

• Name: NvidiaNIM_evo2

• Type: NvidiaNIMTool

• Description: Generate DNA sequences using Evo2-40B via NVIDIA NIM. 40 billion parameter model trained on 9 trillion nucleotides. Input: DNA sequence (ACTG), num_tokens to generate. Output: generated sequence with optional per-nucleotide probabilities. Fast synchronous response.

Parameters:

• sequence (string) (required) Input DNA sequence (A, C, T, G characters only)

• num_tokens (integer) (optional) Number of nucleotides to generate

• temperature (number) (optional) Sampling temperature

• top_k (integer) (optional) Top-k sampling parameter

• top_p (number) (optional) Top-p (nucleus) sampling parameter

• enable_sampled_probs (boolean) (optional) Return per-nucleotide probabilities

• enable_logits (boolean) (optional) Return raw logits

Example Usage:

query = {
    "name": "NvidiaNIM_evo2",
    "arguments": {
        "sequence": "example_value"
    }
}
result = tu.run(query)

NvidiaNIM_genmol (Type: NvidiaNIMTool)

Generate molecules using GenMol via NVIDIA NIM. Input: SMILES/SAFE template with masked regions. …

NvidiaNIM_genmol tool specification

Tool Information:

• Name: NvidiaNIM_genmol

• Type: NvidiaNIMTool

• Description: Generate molecules using GenMol via NVIDIA NIM. Input: SMILES/SAFE template with masked regions. Output: generated molecules with property scores. Use [*{min-max}] syntax for masked fragments. Fast synchronous response.

Parameters:

• smiles (string) (required) SAFE/SMILES template with masked fragments. Use [*{min-max}] for variable-length masks. Example: ‘CC(=O)O[*{5-10}]’

• num_molecules (integer) (optional) Number of molecules to generate (1-1000)

• temperature (number) (optional) Sampling temperature (0.01-10.0). Higher = more diverse

• noise (number) (optional) Noise level for generation (0-2.0)

• step_size (integer) (optional) Step size for sampling

• scoring (string) (optional) Property to score generated molecules

Example Usage:

query = {
    "name": "NvidiaNIM_genmol",
    "arguments": {
        "smiles": "example_value"
    }
}
result = tu.run(query)

NvidiaNIM_maisi (Type: NvidiaNIMTool)

Generate synthetic CT images with segmentation masks using MAISI via NVIDIA NIM. 3D Latent Diffus…

NvidiaNIM_maisi tool specification

Tool Information:

• Name: NvidiaNIM_maisi

• Type: NvidiaNIMTool

• Description: Generate synthetic CT images with segmentation masks using MAISI via NVIDIA NIM. 3D Latent Diffusion Model for medical imaging data augmentation. Input: body_region, anatomy_list, controllable_anatomy_size, output_size. Returns ZIP file with NIfTI/NRRD images and labels. Long-running async operation.

Parameters:

• num_output_samples (integer) (required) Number of CT images to generate

• body_region (array) (required) Body regions to generate

• anatomy_list (array) (optional) Specific anatomical structures to include (e.g., liver, spleen, kidney)

• controllable_anatomy_size (array) (optional) Control anatomy sizes as [[name, ratio], …]. Example: [[‘hepatic tumor’, 0.3], [‘liver’, 0.5]]

• output_size (array) (optional) Output volume size [x, y, z]. x,y: 256-512, z: 128-768

• spacing (array) (optional) Voxel spacing [x, y, z] in mm

• image_output_ext (string) (optional) Output image format

• label_output_ext (string) (optional) Output segmentation format

• pre_signed_url (string) (optional) Optional pre-signed URL to upload results to

Example Usage:

query = {
    "name": "NvidiaNIM_maisi",
    "arguments": {
        "num_output_samples": 10,
        "body_region": ["item1", "item2"]
    }
}
result = tu.run(query)

NvidiaNIM_molmim (Type: NvidiaNIMTool)

Controlled molecule generation using MolMIM via NVIDIA NIM. Generates molecules with optimized pr…

NvidiaNIM_molmim tool specification

Tool Information:

• Name: NvidiaNIM_molmim

• Type: NvidiaNIMTool

• Description: Controlled molecule generation using MolMIM via NVIDIA NIM. Generates molecules with optimized properties based on a reference molecule. Input: smi (SMILES string), num_molecules, algorithm (CMA-ES, GA, etc.). Output: generated molecules with property scores. Fast synchronous response.

Parameters:

• smi (string) (required) Reference SMILES string for molecule generation

• num_molecules (integer) (optional) Number of molecules to generate (1-100)

• algorithm (string) (optional) Optimization algorithm: CMA-ES (Covariance Matrix Adaptation Evolution Strategy) or GA (Genetic Algorithm)

Example Usage:

query = {
    "name": "NvidiaNIM_molmim",
    "arguments": {
        "smi": "example_value"
    }
}
result = tu.run(query)

NvidiaNIM_msa_search (Type: NvidiaNIMTool)

GPU-accelerated multiple sequence alignment search using ColabFold/MMseqs2 via NVIDIA NIM. Input:…

NvidiaNIM_msa_search tool specification

Tool Information:

• Name: NvidiaNIM_msa_search

• Type: NvidiaNIMTool

• Description: GPU-accelerated multiple sequence alignment search using ColabFold/MMseqs2 via NVIDIA NIM. Input: protein sequence. Output: MSA alignments in a3m/fasta format. Use for structure prediction pipelines. Long-running async operation.

Parameters:

• sequence (string) (required) Protein sequence to search (1-4096 amino acids)

• e_value (number) (optional) E-value threshold for sequence inclusion

• iterations (integer) (optional) Number of search iterations for sensitivity

• output_alignment_formats (array) (optional) Output alignment formats

• databases (array) (optional) Specific databases to search

• max_msa_sequences (integer) (optional) Maximum sequences in output MSA (max 10000)

Example Usage:

query = {
    "name": "NvidiaNIM_msa_search",
    "arguments": {
        "sequence": "example_value"
    }
}
result = tu.run(query)

NvidiaNIM_openfold2 (Type: NvidiaNIMTool)

Predict protein structure from sequence and MSA using OpenFold2 via NVIDIA NIM. PyTorch reimpleme…

NvidiaNIM_openfold2 tool specification

Tool Information:

• Name: NvidiaNIM_openfold2

• Type: NvidiaNIMTool

• Description: Predict protein structure from sequence and MSA using OpenFold2 via NVIDIA NIM. PyTorch reimplementation of AlphaFold2 with improved speed. Input: sequence plus pre-computed MSA alignments in a3m format. Long-running async operation.

Parameters:

• sequence (string) (required) Amino acid sequence to predict

• alignments (object) (optional) MSA alignments by database. Format: {db_name: {a3m: {alignment: str, format: ‘a3m’}}}

• selected_models (array) (optional) Model indices to use for prediction (1-5)

Example Usage:

query = {
    "name": "NvidiaNIM_openfold2",
    "arguments": {
        "sequence": "example_value"
    }
}
result = tu.run(query)

NvidiaNIM_openfold3 (Type: NvidiaNIMTool)

Predict biomolecular complex structure (proteins, DNA, RNA, ligands) using OpenFold3 via NVIDIA N…

NvidiaNIM_openfold3 tool specification

Tool Information:

• Name: NvidiaNIM_openfold3

• Type: NvidiaNIMTool

• Description: Predict biomolecular complex structure (proteins, DNA, RNA, ligands) using OpenFold3 via NVIDIA NIM. AlphaFold3-style prediction for complete biological assemblies. Input: molecules (proteins with MSA, DNA, RNA, ligands). Long-running async operation.

Parameters:

• inputs (array) (required) List of input specifications for structure prediction

Example Usage:

query = {
    "name": "NvidiaNIM_openfold3",
    "arguments": {
        "inputs": ["item1", "item2"]
    }
}
result = tu.run(query)

NvidiaNIM_proteinmpnn (Type: NvidiaNIMTool)

Design protein sequences for a given backbone structure using ProteinMPNN via NVIDIA NIM (inverse…

NvidiaNIM_proteinmpnn tool specification

Tool Information:

• Name: NvidiaNIM_proteinmpnn

• Type: NvidiaNIMTool

• Description: Design protein sequences for a given backbone structure using ProteinMPNN via NVIDIA NIM (inverse folding). Input: PDB structure text. Output: designed sequences in Multi-FASTA format with log-probabilities. Fast synchronous response.

Parameters:

• input_pdb (string) (required) PDB format text of the backbone structure (ATOM records)

• ca_only (boolean) (optional) Use only CA atoms for design

• use_soluble_model (boolean) (optional) Use model trained on soluble proteins

• sampling_temp (array) (optional) Sampling temperatures (0.1-0.3 recommended for high quality)

• num_seq_per_target (integer) (optional) Number of sequences to generate

Example Usage:

query = {
    "name": "NvidiaNIM_proteinmpnn",
    "arguments": {
        "input_pdb": "example_value"
    }
}
result = tu.run(query)

NvidiaNIM_rfdiffusion (Type: NvidiaNIMTool)

De novo protein design using RFdiffusion via NVIDIA NIM. Generate novel protein structures, binde…

NvidiaNIM_rfdiffusion tool specification

Tool Information:

• Name: NvidiaNIM_rfdiffusion

• Type: NvidiaNIMTool

• Description: De novo protein design using RFdiffusion via NVIDIA NIM. Generate novel protein structures, binders, or scaffold motifs. Input: contigs DSL (e.g., ‘A20-60/0 50-100’), input_pdb (required for scaffolding), hotspot_res for binding sites. Long-running async operation.

Parameters:

• contigs (string) (required) Contig specification DSL. Format: ‘ChainStart-End/gap length’. Example: ‘A20-60/0 50-100’ keeps residues A20-60 and generates 50-100 new residues

• input_pdb (string) (required) PDB structure for scaffolding/binder design (ATOM records only). Required by the API.

• hotspot_res (array) (optional) Hotspot residues for binder design (e.g., [‘A50’, ‘A51’, ‘A52’])

• diffusion_steps (integer) (optional) Number of diffusion steps (15-50)

• random_seed (integer) (optional) Random seed for reproducibility

Example Usage:

query = {
    "name": "NvidiaNIM_rfdiffusion",
    "arguments": {
        "contigs": "example_value",
        "input_pdb": "example_value"
    }
}
result = tu.run(query)

NvidiaNIM_vista3d (Type: NvidiaNIMTool)

3D medical image segmentation using VISTA-3D via NVIDIA NIM. Segment organs and structures from C…

NvidiaNIM_vista3d tool specification

Tool Information:

• Name: NvidiaNIM_vista3d

• Type: NvidiaNIMTool

• Description: 3D medical image segmentation using VISTA-3D via NVIDIA NIM. Segment organs and structures from CT scans. Input: image URL (NIfTI/NRRD), class prompts, optional point prompts for interactive refinement. Returns ZIP file with segmentation masks. Synchronous response.

Parameters:

• image (string) (required) URL to NIfTI or NRRD CT image file

• prompts (object) (optional) Segmentation prompts: classes and optional point coordinates

Example Usage:

query = {
    "name": "NvidiaNIM_vista3d",
    "arguments": {
        "image": "example_value"
    }
}
result = tu.run(query)

Navigation

• Available Tools - Back to Tools Overview

• Tool Loading Tutorial - Loading Local Tools