# NemoClaw Community

> A catalog of constrained, inspectable agent recipes, field demos, and developer tools built with NemoClaw.

Use the industry and category fields below to select an example. Requirements are short summaries; read the linked source guide before running an example.

## Catalog data

- [Browse the catalog](https://nvidia.github.io/nemoclaw-community/)
- [Machine-readable catalog](https://nvidia.github.io/nemoclaw-community/catalog.json)

## Examples

- [Agentic AI Learning Path](https://nvidia.github.io/nemoclaw-community/examples/recipes/nvidia/agentic-ai-learning-path/)
  - Description: Helps learners work through the seven-module Build an Agent workshop in JupyterLab with an AI tutor that explains concepts, offers graduated hints, and checks progress inside an OpenShell sandbox.
  - Category: NVIDIA Recipes
  - Industry: 🎓 Academia/Education
  - Requirements: Linux · Docker · OpenShell · inference, NVIDIA, and Tavily API keys · Slack or Outlook · CPU-first; module 4 training is read-through
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/nvidia/agentic-ai-learning-path/README.md)
  - NemoClaw version: N/A
  - Harness: Hermes
  - Harness version: 0.20.0
  - OpenShell version: Unpinned
  - Stack verification: unpinned
  - Maintenance: Current
  - Upstream: [Project](<https://github.com/brevdev/workshop-build-an-agent>)
- [Developer Community Chief of Staff](https://nvidia.github.io/nemoclaw-community/examples/recipes/nvidia/developer-community-chief-of-staff/)
  - Description: Helps developer community leaders align priorities with demand by turning available GitHub, GitLab, forum, Slack, email, and web signals into evidence-grounded briefs, gaps, and follow-up recommendations.
  - Category: NVIDIA Recipes
  - Industry: ✨ Other
  - Requirements: Single Linux host · Docker · OpenShell · inference provider API key · Slack or Outlook
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/nvidia/developer-community-chief-of-staff/README.md)
  - NemoClaw version: N/A
  - Harness: Hermes
  - Harness version: 0.20.0
  - OpenShell version: 0.0.85
  - Stack verification: unconfirmed
  - Maintenance: Current
- [Kubernetes GPU Autoscaling](https://nvidia.github.io/nemoclaw-community/examples/recipes/nvidia/kubernetes-gpu-autoscaling/)
  - Description: Helps Kubernetes operators match GPU inference capacity to demand by pairing a CPU-only OpenShell sandbox with Ollama replicas that scale on utilization or latency and return to one after load.
  - Category: NVIDIA Recipes
  - Industry: ☁️ Cloud Services
  - Requirements: Kubernetes 1.25+ · Helm 3 · NVIDIA GPU Operator/DCGM · Metrics Server · Docker Buildx + registry · OpenShell + Agent Sandbox CRDs pinned in versions.env · OIDC or acknowledged isolated-eval exception · experimental
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/nvidia/kubernetes-gpu-autoscaling/README.md)
  - NemoClaw version: v0.0.104
  - Harness: OpenClaw
  - Harness version: 2026.7.1
  - OpenShell version: 0.0.85
  - Stack verification: unconfirmed
  - Maintenance: Current
- [Memory-Driven Chief of Staff](https://nvidia.github.io/nemoclaw-community/examples/recipes/nvidia/memory-driven-chief-of-staff/)
  - Description: Builds a revisable local memory from email and Slack, then ranks obligations against the user's priorities while preserving pins and ignores without changing source systems.
  - Category: NVIDIA Recipes
  - Industry: ✨ Other
  - Requirements: Python 3.10+ · scheduled use: a Linux NemoHermes sandbox, Hermes 0.19+, and an inference provider API key · provider setup helpers: Linux/WSL · Slack and Microsoft Graph collectors optional
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/nvidia/memory-driven-chief-of-staff/README.md)
  - NemoClaw version: Unpinned
  - Harness: Hermes
  - Harness version: >=0.19.0
  - OpenShell version: Unpinned
  - Stack verification: unpinned
  - Maintenance: Current
- [NV Tech Assistant](https://nvidia.github.io/nemoclaw-community/examples/recipes/nvidia/nv-tech-assistant/)
  - Description: Helps developers choose and troubleshoot NVIDIA technologies by searching allowlisted documentation, repositories, model catalogs, and forums for current, evidence-linked answers.
  - Category: NVIDIA Recipes
  - Industry: 🖥️ Hardware/Semiconductor
  - Requirements: Docker · NemoClaw · NVIDIA Endpoints API key · optional Brave Search · no offline corpus
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/nvidia/nv-tech-assistant/README.md)
  - NemoClaw version: Unpinned
  - Harness: OpenClaw
  - Harness version: Unpinned
  - OpenShell version: Unpinned
  - Stack verification: unpinned
  - Maintenance: Current
- [Payment Operations Hermes Assistant](https://nvidia.github.io/nemoclaw-community/examples/recipes/nvidia/payment-ops-hermes/)
  - Description: Helps payment operators screen synthetic outbound payments, explain holds, and prepare review packets while OpenShell keeps final release authority with a human outside the Hermes sandbox.
  - Category: NVIDIA Recipes
  - Industry: 💳 Financial Services
  - Requirements: Linux · Docker Compose · OpenShell 0.0.53 · OpenAI-compatible inference API key · synthetic data only
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/nvidia/payment-ops-hermes/README.md)
  - NemoClaw version: N/A
  - Harness: Hermes
  - Harness version: 0.14.0
  - OpenShell version: 0.0.53
  - Stack verification: unconfirmed
  - Maintenance: Current
- [PR Review Advisor](https://nvidia.github.io/nemoclaw-community/examples/recipes/nvidia/pr-review-advisor/)
  - Description: Helps maintainers review exact pull request heads through staged, repository-aware Hermes analysis, producing attested JSON and Markdown findings for inspection before optional publication.
  - Category: NVIDIA Recipes
  - Industry: ✨ Other
  - Requirements: Linux + Docker · OpenShell 0.0.85 · Node.js 22.19+ · inference provider API key or existing provider · GitHub.com self-hosted runner for the generated workflow
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/nvidia/pr-review-advisor/README.md)
  - NemoClaw version: N/A
  - Harness: Hermes
  - Harness version: 0.18.0
  - OpenShell version: 0.0.85
  - Stack verification: unconfirmed
  - Maintenance: Current
- [PR Test Case Assistant](https://nvidia.github.io/nemoclaw-community/examples/recipes/nvidia/pr-test-case-assistant/)
  - Description: Helps quality engineers turn public GitHub pull request descriptions and bounded diffs into Slack briefs and proposed, unexecuted feature test cases with source evidence.
  - Category: NVIDIA Recipes
  - Industry: ✨ Other
  - Requirements: NemoClaw + Docker · Slack app with Socket Mode · Slack bot + app access tokens · inference provider API key · public GitHub only
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/nvidia/pr-test-case-assistant/README.md)
  - NemoClaw version: Unpinned
  - Harness: OpenClaw
  - Harness version: Unpinned
  - OpenShell version: Unpinned
  - Stack verification: unpinned
  - Maintenance: Current
- [Sandboxed Hermes Bot Team](https://nvidia.github.io/nemoclaw-community/examples/recipes/nvidia/sandboxed-hermes-bots/)
  - Description: A team of Hermes bots you talk to from Hermes Desktop, one NemoClaw sandbox each. A bot reaches only what its policy names, and when it needs something it cannot reach, it asks a teammate.
  - Category: NVIDIA Recipes
  - Industry: ✨ Other
  - Requirements: Linux host or macOS with Colima · Docker · OpenShell and NemoClaw · Hermes 0.21 on the host · OpenAI-compatible inference endpoint · optional GPU for the video example
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/nvidia/sandboxed-hermes-bots/README.md)
  - NemoClaw version: Unpinned
  - Harness: Hermes
  - Harness version: 0.21.0
  - OpenShell version: 0.0.101
  - Stack verification: unpinned
  - Maintenance: Current
  - Upstream: [Project](<https://github.com/PicoNVIDIA/sandboxed-bots>)
- [Video Search and Summarization](https://nvidia.github.io/nemoclaw-community/examples/recipes/nvidia/video-search-and-summarization/)
  - Description: Helps video analysts and engineers deploy and operate NVIDIA VSS profiles by chat, using a sandboxed agent that runs the Compose deployment through a host-side MCP server and reports the result.
  - Category: NVIDIA Recipes
  - Industry: ✨ Other
  - Requirements: Linux GPU host · Docker with the NVIDIA runtime · Python 3.11+ for the notebooks · NGC API key · agent model provider · VSS repository develop branch · large container image and model pulls
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/nvidia/video-search-and-summarization/README.md)
  - NemoClaw version: Unpinned
  - Harness: Unknown
  - Harness version: Unknown
  - OpenShell version: Unpinned
  - Stack verification: unknown
  - Maintenance: Current
- [x402 Payment Gate](https://nvidia.github.io/nemoclaw-community/examples/recipes/partners/bluetier/x402-payment-gate/)
  - Description: Demonstrates a maker-checker gate for x402 payments: a sandboxed agent submits intents, while a host-side Blackwall verdict controls mock signing and settlement before any signature exists.
  - Category: Partner Recipes
  - Industry: 💳 Financial Services
  - Requirements: Python 3 · Docker + OpenShell for sandbox checks · Blackwall service · simulated signing/no real money
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/partners/bluetier/x402-payment-gate/README.md)
  - NemoClaw version: N/A
  - Harness: Hermes
  - Harness version: 0.18.0
  - OpenShell version: >=0.0.85
  - Stack verification: unpinned
  - Maintenance: Current
- [Retail Assistant](https://nvidia.github.io/nemoclaw-community/examples/recipes/partners/hpe/retail-assistant/)
  - Description: Helps store employees check inventory and sales or request transfers and reorders through role-aware Telegram conversations scoped to their assigned store.
  - Category: Partner Recipes
  - Industry: 🛍️ Retail/Consumer Packaged Goods
  - Requirements: Telegram bot · tool-calling OpenAI-compatible LLM · Docker Compose or Kubernetes + Helm · demo identity mapping
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/partners/hpe/retail-assistant/README.md)
  - NemoClaw version: Unpinned
  - Harness: OpenClaw
  - Harness version: Unpinned
  - OpenShell version: Unpinned
  - Stack verification: unpinned
  - Maintenance: Current
- [Shrike Security Action Governance](https://nvidia.github.io/nemoclaw-community/examples/recipes/partners/shrike/shrike-security/)
  - Description: Adds defense-in-depth action governance through an in-sandbox hook that checks each OpenClaw tool call against server-side Shrike policy and blocks prohibited or approval-required calls.
  - Category: Partner Recipes
  - Industry: ✨ Other
  - Requirements: NemoClaw/OpenShell · Node.js + npm · inference provider · Shrike API key · in-sandbox defense-in-depth only
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/partners/shrike/shrike-security/README.md)
  - NemoClaw version: >=v0.0.76
  - Harness: OpenClaw
  - Harness version: Unpinned
  - OpenShell version: Unpinned
  - Stack verification: unpinned
  - Maintenance: Current
- [Watchtower](https://nvidia.github.io/nemoclaw-community/examples/recipes/partners/tavily/watchtower/)
  - Description: Tracks what changed across chosen web topics and why it matters, producing scheduled, deduplicated Markdown digests and JSON changelogs with source citations.
  - Category: Partner Recipes
  - Industry: ✨ Other
  - Requirements: Docker · NemoClaw · Tavily and inference provider API keys · one-time operator.admin approval
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/partners/tavily/watchtower/README.md)
  - NemoClaw version: Unpinned
  - Harness: OpenClaw
  - Harness version: Unpinned
  - OpenShell version: Unpinned
  - Stack verification: unpinned
  - Maintenance: Current
- [Axe A11y Browser Auditor](https://nvidia.github.io/nemoclaw-community/examples/recipes/community/axe-a11y-browser-auditor/)
  - Description: Helps web teams find WCAG issues through a dedicated sandbox and host-side real Chrome, returning rule-level findings with optional screenshots, PDFs, recordings, and network traces.
  - Category: Community Recipes
  - Industry: 🌐 Consumer Internet
  - Requirements: Docker 24+ with Compose · linux/amd64 or Docker Desktop emulation · dedicated OpenShell/NemoClaw sandbox
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/community/axe-a11y-browser-auditor/README.md)
  - NemoClaw version: Unpinned
  - Harness: OpenClaw
  - Harness version: Unpinned
  - OpenShell version: Unpinned
  - Stack verification: unpinned
  - Maintenance: Current
- [Deep Research Worker](https://nvidia.github.io/nemoclaw-community/examples/recipes/community/deep-research-worker/)
  - Description: Adds asynchronous research to a sandbox, so users can queue long-running jobs on a host-side worker, monitor persistent task state, and return later for text or JSON results.
  - Category: Community Recipes
  - Industry: ✨ Other
  - Requirements: Docker Compose · dedicated OpenShell/NemoClaw sandbox · OpenAI-compatible inference endpoint + API key · no readiness guarantee
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/recipes/community/deep-research-worker/README.md)
  - NemoClaw version: Unpinned
  - Harness: LangChain Deep Agents
  - Harness version: 0.7.5
  - OpenShell version: Unpinned
  - Stack verification: unpinned
  - Maintenance: Current
- [Build-a-Claw Tutorial](https://nvidia.github.io/nemoclaw-community/examples/demos/field/build-a-claw-tutorial/)
  - Description: Guides DGX Spark users through serving local multimodal models with llama.cpp, connecting an agent harness, and trying coding, vision, browser, messaging, and speech workflows.
  - Category: NVIDIA Field Demos
  - Industry: 🎓 Academia/Education
  - Requirements: NVIDIA DGX Spark · Ubuntu with sudo, CUDA build tools, Node.js 24.15.0, Python 3, internet, and storage for large GGUF downloads · optional external accounts and device permissions
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/demos/field/build-a-claw-tutorial/README.md)
  - NemoClaw version: N/A
  - Harness: OpenClaw
  - Harness version: 2026.7.1-2
  - OpenShell version: N/A
  - Stack verification: unconfirmed
  - Maintenance: Current
  - Collections: Build-a-Claw
- [DGX Station Blender and Omniverse](https://nvidia.github.io/nemoclaw-community/examples/demos/field/blender-omniverse-dgx-station/)
  - Description: Lets users direct a specialized Hermes agent on DGX Station across Blender and NVIDIA Omniverse workflows, producing scene edits, OVRTX renders, native OVPhysX simulations, and replay evidence.
  - Category: NVIDIA Field Demos
  - Industry: 🎬 Media & Entertainment
  - Requirements: DGX Station · Ubuntu 24.04 ARM64 · GB300 + RTX GPU · Blender 5.1.x + OVRTX/OVPhysX · local vLLM with Nemotron 3 Ultra · Hugging Face read access token
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/demos/field/blender-omniverse-dgx-station/README.md)
  - NemoClaw version: v0.0.83
  - Harness: Hermes
  - Harness version: 0.18.0
  - OpenShell version: 0.0.72
  - Stack verification: unconfirmed
  - Maintenance: Current
- [Agent Memory Benchmark](https://nvidia.github.io/nemoclaw-community/examples/tools/agent-memory-benchmark/)
  - Description: Measures memory built from synthetic email and chat, asks 186 questions on one corpus and 96 on a second, and reports accuracy by question type with ingest and answer token costs.
  - Category: Developer Tools
  - Industry: ✨ Other
  - Requirements: Python 3.9+ · pytest and tesseract for offline checks · endpoint and adapter for live scoring
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/tools/agent-memory-benchmark/README.md)
  - NemoClaw version: N/A
  - Harness: N/A
  - Harness version: N/A
  - OpenShell version: N/A
  - Stack verification: not-applicable
  - Maintenance: Current
- [Harness Engineering Playground](https://nvidia.github.io/nemoclaw-community/examples/tools/harness-engineering-playground/)
  - Description: Provides an experimental loop for tuning DeepAgents harness profiles against behavioral evaluations, keeping fixes that pass verification and rolling back rejected edits.
  - Category: Developer Tools
  - Industry: ✨ Other
  - Requirements: Python 3.11+ · uv · DeepAgents submodule + extra · proposer and target-model API keys · experimental/not for production
  - Lifecycle: Active
  - Source: [README](https://github.com/NVIDIA/nemoclaw-community/blob/main/examples/tools/harness-engineering-playground/README.md)
  - NemoClaw version: N/A
  - Harness: LangChain Deep Agents
  - Harness version: 0.6.12
  - OpenShell version: N/A
  - Stack verification: unconfirmed
  - Maintenance: Current