This agent follows the Nebius Serverless AI cookbook BioNeMo recipe. It runs a CPU agent endpoint backed by TokenFactory and calls a GPU-backed BioNeMo-compatible service endpoint for protein embeddings, structure prediction, retrieval, MD metadata, genomics generation, and chat.
GPU B200 SXM A · 1 GPU · 20 vCPU · 224 GB
CPU D3 · 4 vCPU · 16 GB
GPU B200 SXM A · 1 GPU · 20 vCPU · 224 GB, CPU D3 · 4 vCPU · 16 GB
BioNeMo-compatible service paths exposed to the agent.
Routes nonclinical biomedical text generation and educational summarization requests.
Routes public or synthetic protein sequences to an ESM-2-style embedding service.
Prepares Boltz2-style nonclinical structure prediction calls.
Routes retrieval and literature-search embedding requests.
Returns OpenMM-style molecular dynamics timing and artifact metadata.
Current Forge validation project exposed only gpu-b200-sxm-a for Serverless GPU jobs on 2026-06-18.
| Platform | Preset | GPU | Status | Tested |
|---|---|---|---|---|
| gpu-b200-sxm-a | 1gpu-20vcpu-224gb | 1 × NVIDIA B200 · 179 GB | supported | 2026-06-18T14:40:27Z |
| gpu-b200-sxm-a | 8gpu-160vcpu-1792gb | 8 × NVIDIA B200 · 179 GB | supported | 2026-06-18T14:46:14Z |
Matrix scope: service smoke only. It confirms the service image starts, sees the requested GPU shape, and exercises all advertised demo skills.
Create the GPU service first, then create the CPU agent endpoint against it.
Step 1
GPU-backed service endpoint that exposes the BioNeMo-compatible skills the agent calls.
Step 2
CPU orchestration endpoint that serves the NeMo Agent Toolkit workflow and calls the service backend.
Routes synthetic DNA/RNA generation demos.
Required inputs
