NVIDIA recipe
Agentic AI Learning Path
Runs the seven-module Build an Agent workshop in an OpenShell sandbox with an AI tutor.
- Fit
- Linux + Docker + OpenShell · API keys + Slack or Outlook · CPU-first; module 4 training read-through
Example catalog
Start with the kind of example you need.
Browse NVIDIA, partner, and community recipes, field demos, launchables, and developer tools.
Example catalog
NVIDIA recipes
Reusable agent recipes authored or maintained by NVIDIA. Placement does not state a support or maturity level.
NVIDIA recipe
Developer Community Chief of Staff
Turns community signals into operating briefs and follow-up recommendations.
- Fit
- Docker · Slack/Outlook · single host
NVIDIA recipe
Kubernetes GPU Autoscaling
Runs NemoClaw on Kubernetes with GPU-backed Ollama autoscaling.
- Fit
- Kubernetes · NVIDIA GPU · experimental
NVIDIA recipe
Memory-Driven Chief of Staff
Maintains and re-ranks a revisable local record of email and Slack inputs without writing back.
- Fit
- Python 3.10+ · Hermes 0.19+ · connectors not included
NVIDIA recipe
NV Tech Assistant
Answers NVIDIA technical questions with citations from allowlisted sources.
- Fit
- Docker · NVIDIA Endpoints API key · no offline corpus
NVIDIA recipe
Payment Operations Hermes Assistant
Demonstrates payment screening, evidence preparation, and a platform-enforced human release boundary.
- Fit
- Docker Compose · OpenShell 0.0.53 · synthetic data
NVIDIA recipe
PR Review Advisor
Reviews exact pull request heads with Hermes, produces attested artifacts, and leaves publication to a maintainer.
- Fit
- OpenShell · GitHub.com only · self-hosted runner
Partner recipes
Reusable agent recipes contributed by named partner organizations. Partner attribution remains visible on each card.
Partner recipe · BlueTier
x402 Payment Gate
Screens x402 payment intents through a host-side maker/checker gate before signing or settlement.
- Fit
- Python 3 + Docker · Blackwall · no real money
Partner recipe · HPE
Retail Assistant
Provides role-aware retail operations through Telegram with Docker Compose or Helm deployment paths.
- Fit
- Telegram · Docker or Kubernetes · demo identity mapping
Partner recipe · Shrike Security
Shrike Security Action Governance
Applies Shrike decisions before agent tool calls run.
- Fit
- NemoClaw/OpenShell · Shrike API · not an independent security boundary
Partner recipe · Tavily
Watchtower
Runs scheduled, cited web monitoring with deduplication and auditable outputs.
- Fit
- Docker + NemoClaw · Tavily · administrator approval
Community recipes
Reusable agent recipes contributed independently without formal organizational provenance.
Community recipe
Axe A11y Browser Auditor
Audits web pages for accessibility and captures screenshots, PDFs, and network traces.
- Fit
- Docker 24+ · linux/amd64 · dedicated sandbox
Community recipe
Deep Research Worker
Queues long-running research from a sandbox to a host-side DeepAgents worker.
- Fit
- Docker Compose · dedicated sandbox · no readiness guarantee
NVIDIA field demos
Bounded NVIDIA demonstrations for named hardware and software environments.
NVIDIA field demo
DGX Station Blender and Omniverse
Controls Blender on DGX Station for OVRTX rendering and OVPhysX simulation.
- Fit
- Ubuntu 24.04 ARM64 · Blender 5.1.x · GB300 + RTX PRO 6000 validation
Launchables
Provisioning and onboarding paths for a specific environment.
Launchable · Brev
Hermes
Takes a fresh Brev CPU instance to a NemoClaw-managed Hermes sandbox through a notebook.
- Fit
- Brev CPU · NVIDIA Build API key · best-effort support
Developer tools
Standalone development and evaluation utilities, separate from deployed agent blueprints.
Developer tool
Agent Memory Benchmark
Measures the memory a system builds from a corpus: it feeds a system synthetic email and chat, asks 186 questions on one corpus and 96 on a second, and reports accuracy per question type alongside ingest and per-answer token cost.
- Fit
- Python 3.9+ · offline self-test · endpoint optional
Developer tool
Harness Engineering Playground
Provides an experimental command-line interface for evaluation-driven harness profile optimization.
- Fit
- Python 3.11+ · DeepAgents · not for production