# Set Up AI Runway on AKS

AI Runway setup requires coordinated cluster, GPU, provider, and model checks. This skill guides each AKS stage with confirmation before installation or deployment.

## Install

```bash
npx skillstore add microsoft/airunway-aks-setup
```

## Metadata

- Status: approved
- Slug: microsoft-airunway-aks-setup
- Version: 1.0.1
- Author version: 1.0.1
- Skillstore revision: r2
- Version status: valid
- Tree hash: 3a859beeedcb4467874f9ad35a1acb36505c3af19884af62cb1737ff7253b673
- Author: microsoft
- GitHub username: microsoft
- License: MIT
- Repository: https://github.com/microsoft/azure-skills/tree/main/.github/plugins/azure-skills/skills/airunway-aks-setup/
- Ref: ebdfe608f5de2b66ff37ab4af12af8ac4f5e8006
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, filesystem
- Quality score: 71
- Public page: https://skillstore.pages.dev/skills/microsoft-airunway-aks-setup
- Manifest: https://skillstore.pages.dev/api/skills/microsoft-airunway-aks-setup/manifest

## Capabilities

- Verifies required CLI tools, the active Kubernetes context, node inventory, and detected NVIDIA GPU capacity.
- Checks AI Runway controller and custom resource status before guiding confirmed installation through repository Make targets.
- Matches detected GPUs to VRAM, data type, attention backend, and model sizing guidance.
- Recommends KAITO, Dynamo, or KubeRay based on hardware and serving needs.
- Creates Kubernetes model resources, monitors readiness, and provides rollback checks for failed deployments.
- Retrieves the model endpoint and guides a basic chat completion smoke test.

## Use Cases

- Bootstrap an existing AKS cluster: Verify the cluster, install AI Runway components with confirmation, and deploy a compatible starter model.
- Match models to GPU capacity: Assess detected accelerators, identify compatibility limits, and select a provider and model that fit available VRAM.
- Resume or recover setup: Start from a named setup step, inspect component status, and follow targeted readiness or rollback guidance.

## Prompt Templates

### Verify an AKS cluster

```
Verify my current AKS cluster for AI Runway. Report tools, context, nodes, GPUs, and blockers. Do not install anything.
```

### Complete guided setup

```
Set up AI Runway on my existing AKS cluster. Stop before each installation or deployment and explain the expected cluster changes.
```

### Choose and deploy a model

```
Assess my GPU capacity, recommend a compatible provider and starter model, then deploy after confirmation. Monitor readiness and explain any failure.
```

### Resume with an advanced provider

```
Resume at step 4 on my AKS cluster. Compare Dynamo and KubeRay for high-throughput serving, verify the selected image, and request confirmation before deployment.
```

## Limitations

- Requires an existing AKS cluster, working kubeconfig, and local access to kubectl, make, curl, and the AI Runway repository.
- Does not create AKS clusters, GPU node pools, cloud quotas, ingress, or production security controls.
- Hardware profiles and starter model estimates may not reflect newer GPUs, provider releases, or runtime overhead.
- Model licenses, gated access approvals, image trust, and ongoing cloud costs remain user responsibilities.

## Best Practices

- Verify the active kubeconfig context and namespace before every mutating command.
- Review GPU costs, model licenses, image sources, and expected cluster changes before confirmation.
- Use interactive secret entry, least-privilege access, and immediate temporary file cleanup for gated model tokens.

## Anti Patterns

- Do not skip prerequisite or readiness checks when resuming from a later step.
- Do not place Hugging Face tokens in prompts, shell history, manifests, or shared logs.
- Do not select models from weight size alone; reserve memory for cache and runtime overhead.

## Security Audit

- Audited at: 2026-07-23T17:34:42.383\+00:00
- Summary: All 46 static findings are false positives from Markdown syntax, bundled relative links, or expected administrative command examples. Mutating operations require user confirmation, and no prompt injection or malicious intent was found. A separate medium risk remains because the PowerShell flow stores a token at a predictable plaintext temporary path.

## Stats

- Views: 54
- Downloads: 9
- Favorites: 0
- Popularity score: 0