feat(examples): add Kubernetes GPU autoscaling recipe

THIRD-PARTY-NOTICES

@@ -31,6 +31,14 @@ Image: vllm/vllm-openai
 License: Apache-2.0 and included component licenses
 URL: https://github.com/vllm-project/vllm
 
+Image: node:22-bookworm-slim
+License: MIT and included Debian component licenses
+URL: https://hub.docker.com/_/node
+
+Image: ollama/ollama
+License: MIT and included component licenses
+URL: https://github.com/ollama/ollama
+
 ================================================================================
 Docker Compose Images
 ================================================================================
@@ -215,6 +223,46 @@ Project: NVIDIA/OpenShell installer and runtime
 License: Apache-2.0
 URL: https://github.com/NVIDIA/OpenShell
 
+Project: NVIDIA/NemoClaw source and sandbox base image
+License: Apache-2.0 and included component licenses
+URL: https://github.com/NVIDIA/NemoClaw
+
+Project: Kubernetes SIG Agent Sandbox controller and API
+License: Apache-2.0
+URL: https://github.com/kubernetes-sigs/agent-sandbox
+
+Project: Kubernetes and kubectl
+License: Apache-2.0
+URL: https://github.com/kubernetes/kubernetes
+
+Project: Helm
+License: Apache-2.0
+URL: https://github.com/helm/helm
+
+Project: Canonical MicroK8s distribution
+License: See the project and included component notices
+URL: https://github.com/canonical/microk8s
+
+Project: NVIDIA GPU Operator
+License: Apache-2.0
+URL: https://github.com/NVIDIA/gpu-operator
+
+Project: NVIDIA DCGM Exporter
+License: Apache-2.0
+URL: https://github.com/NVIDIA/dcgm-exporter
+
+Project: Kubernetes Metrics Server
+License: Apache-2.0
+URL: https://github.com/kubernetes-sigs/metrics-server
+
+Project: Prometheus Community Helm Charts (kube-prometheus-stack and prometheus-adapter)
+License: Apache-2.0 and included component licenses
+URL: https://github.com/prometheus-community/helm-charts
+
+Project: Kubernetes ingress-nginx
+License: Apache-2.0
+URL: https://github.com/kubernetes/ingress-nginx
+
 Project: Astral uv installer and package manager
 License: Apache-2.0 OR MIT
 URL: https://github.com/astral-sh/uv
@@ -239,6 +287,10 @@ Model: NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4
 License: NVIDIA Open Model License
 URL: https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4
 
+Model: Llama 3.2 3B
+License: Llama 3.2 Community License Agreement and Acceptable Use Policy
+URL: https://ollama.com/library/llama3.2
+
 Service: GitHub API
 Terms: See provider terms
 URL: https://docs.github.com/

examples/README.md

@@ -11,6 +11,7 @@ again by contributor provenance.
 | Example | Description |
 | --- | --- |
 | [Developer Community Chief of Staff](recipes/nvidia/developer-community-chief-of-staff/README.md) | Synthesizes Slack, Outlook, GitHub, and mirrored community signals into operating briefs, gaps, priorities, and follow-up recommendations. |
+| [Kubernetes GPU Autoscaling](recipes/nvidia/kubernetes-gpu-autoscaling/README.md) | Runs a CPU-only NemoClaw/OpenClaw sandbox through OpenShell on Kubernetes and autoscales authenticated, GPU-backed Ollama inference pods from DCGM utilization. |
 | [NV Tech Assistant](recipes/nvidia/nv-tech-assistant/README.md) | Answers NVIDIA technical questions with cited evidence from allowlisted NVIDIA, GitHub, and arXiv sources. |
 | [Payment Operations Hermes Assistant](recipes/nvidia/payment-ops-hermes/README.md) | Demonstrates constrained payment screening, evidence preparation, and a platform-enforced human release boundary. |
 | [PR Review Advisor](recipes/nvidia/pr-review-advisor/README.md) | Reviews exact pull request heads with a constrained Hermes workflow, produces attested artifacts, and publishes only through a separate maintainer action. |

examples/recipes/nvidia/kubernetes-gpu-autoscaling/.helmignore

@@ -0,0 +1,5 @@
+.DS_Store
+*.swp
+*.bak
+*.tmp
+.git/

examples/recipes/nvidia/kubernetes-gpu-autoscaling/Chart.yaml

@@ -0,0 +1,14 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+apiVersion: v2
+name: nemoclaw-gpu
+description: GPU-backed Ollama inference pods on Kubernetes with GPU utilization HPA
+type: application
+version: 0.1.0
+appVersion: "2026.05.27"
+keywords:
+  - nemoclaw
+  - ollama
+  - gpu
+maintainers:
+  - name: maggiezha

examples/recipes/nvidia/kubernetes-gpu-autoscaling/files/agent-metrics.ts

@@ -0,0 +1,89 @@
+// SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+// SPDX-License-Identifier: Apache-2.0
+//
+// Shared Prometheus helpers for agent /metrics (LLM latency, HTTP counters).
+
+const configuredLlmLatencyWindow = Number(process.env.LLM_LATENCY_WINDOW_SIZE ?? "128");
+const LLM_LATENCY_WINDOW =
+  Number.isSafeInteger(configuredLlmLatencyWindow) && configuredLlmLatencyWindow > 0
+    ? Math.min(configuredLlmLatencyWindow, 10_000)
+    : 128;
+const llmDurationsMs = [];
+let llmDurationSumSec = 0;
+let llmDurationCount = 0;
+let llmRequestsOk = 0;
+let llmRequestsError = 0;
+const llmHistogramBucketsSec = [0.25, 0.5, 1, 2.5, 5, 10, 30, 60, 120, 300];
+const llmHistogramCounts = Array.from({ length: llmHistogramBucketsSec.length + 1 }, () => 0);
+
+export function recordLlmLatency(durationMs, ok) {
+  // Normalize once so the rolling window (p50/p95/avg) and the cumulative
+  // counters/histogram below always agree on the same finite, non-negative value.
+  const normalizedMs = Number.isFinite(durationMs) ? Math.max(0, durationMs) : 0;
+  const sec = normalizedMs / 1000;
+  llmDurationSumSec += sec;
+  llmDurationCount += 1;
+  if (ok) llmRequestsOk += 1;
+  else llmRequestsError += 1;
+
+  llmDurationsMs.push(normalizedMs);
+  if (llmDurationsMs.length > LLM_LATENCY_WINDOW) llmDurationsMs.shift();
+
+  let bucketIdx = llmHistogramBucketsSec.findIndex((bound) => sec <= bound);
+  if (bucketIdx === -1) bucketIdx = llmHistogramBucketsSec.length;
+  for (let i = bucketIdx; i < llmHistogramCounts.length; i += 1) {
+    llmHistogramCounts[i] += 1;
+  }
+}
+
+function percentileMs(sorted, p) {
+  if (!sorted.length) return 0;
+  const idx = Math.ceil(sorted.length * p) - 1;
+  return sorted[Math.max(0, idx)];
+}
+
+function llmLatencySnapshotMs() {
+  if (!llmDurationsMs.length) {
+    return { p50: 0, p95: 0, avg: 0 };
+  }
+  const sorted = [...llmDurationsMs].sort((a, b) => a - b);
+  const sum = sorted.reduce((acc, v) => acc + v, 0);
+  return {
+    p50: percentileMs(sorted, 0.5),
+    p95: percentileMs(sorted, 0.95),
+    avg: sum / sorted.length,
+  };
+}
+
+export function llmMetricsLines() {
+  const { p50, p95, avg } = llmLatencySnapshotMs();
+  const lines = [
+    "# HELP nemoclaw_llm_requests_total Chat/completions proxied to inference backend",
+    "# TYPE nemoclaw_llm_requests_total counter",
+    `nemoclaw_llm_requests_total{result="success"} ${llmRequestsOk}`,
+    `nemoclaw_llm_requests_total{result="error"} ${llmRequestsError}`,
+    "# HELP nemoclaw_llm_request_duration_seconds LLM chat/completions end-to-end proxy latency",
+    "# TYPE nemoclaw_llm_request_duration_seconds histogram",
+  ];
+
+  for (let i = 0; i < llmHistogramBucketsSec.length; i += 1) {
+    lines.push(
+      `nemoclaw_llm_request_duration_seconds_bucket{le="${llmHistogramBucketsSec[i]}"} ${llmHistogramCounts[i]}`,
+    );
+  }
+  lines.push(
+    `nemoclaw_llm_request_duration_seconds_bucket{le="+Inf"} ${llmHistogramCounts[llmHistogramCounts.length - 1]}`,
+    `nemoclaw_llm_request_duration_seconds_sum ${llmDurationSumSec}`,
+    `nemoclaw_llm_request_duration_seconds_count ${llmDurationCount}`,
+    "# HELP nemoclaw_llm_latency_p50_milliseconds Rolling p50 LLM latency (recent window)",
+    "# TYPE nemoclaw_llm_latency_p50_milliseconds gauge",
+    `nemoclaw_llm_latency_p50_milliseconds ${Math.round(p50)}`,
+    "# HELP nemoclaw_llm_latency_p95_milliseconds Rolling p95 LLM latency (recent window)",
+    "# TYPE nemoclaw_llm_latency_p95_milliseconds gauge",
+    `nemoclaw_llm_latency_p95_milliseconds ${Math.round(p95)}`,
+    "# HELP nemoclaw_llm_latency_avg_milliseconds Rolling average LLM latency (recent window)",
+    "# TYPE nemoclaw_llm_latency_avg_milliseconds gauge",
+    `nemoclaw_llm_latency_avg_milliseconds ${Math.round(avg)}`,
+  );
+  return lines;
+}