NVIDIA AI-Q Deployment
Deploy, verify, and operate NVIDIA AI-Q Blueprint infrastructure.
The skill explicitly keeps credentials in deploy/.env, performs presence-only checks, forbids printing secrets, requires confirmation for destructive cleanup, and limits REQUIRE_AUTH=false to trusted local single-user validation. However, the default skill-backend path may still add REQUIRE_AUTH=false, remote cloning lacks upfront user confirmation, the workflow requests broad Git/network/runtime access, and rollback is mostly delegated to referenced documents. Deducted 7 points.
The instructions provide prerequisites, expected outputs, health checks, troubleshooting for ports, credentials, configuration, and cleanup, plus controlled failure guidance. However, this is a static-only review; several key referenced files are not included, and there is no directly reproducible committed test suite covering the key paths. A conservative static score of 8 is used, with 12 points deducted.
Trigger wording, deployment modes, prerequisites, handoff boundary to aiq-research, and version-compatibility rules are reasonably clear. The skill provides no Chinese-language interaction guidance or mainland-China reachability assessment, while core operation depends on GitHub, NVIDIA-hosted endpoints, and search providers. Deducted 4 points.
The documentation is layered and includes references, examples, limitations, version 2.1.0, author, and GitHub provenance; the repository also supplies licensing, sync-maintenance, and security-reporting context. However, the skill declares Apache-2.0 while the repository describes Apache-2.0 plus CC-BY-4.0, and there is no clear changelog, maintenance commitment, or complete FAQ. Deducted 5 points.
The goal is concrete: deploy and basically validate AI-Q, then hand off a server URL. Examples, health checks, mode routing, and evaluation tasks cover the core workflow. Static review cannot confirm command compatibility with the target Blueprint revision, and referenced runtime evidence cannot be independently verified, so the score remains below the static ceiling with 9 points deducted.
BENCHMARK.md and evals.json provide auditable leads, expected behavior, and reported evaluation metrics. However, the results are repository-authored claims without verifiable CI, committed key-path tests, or independent reproduction, and no commands were executed in this review. Deducted 6 points.
- The default Skill backend may set REQUIRE_AUTH=false; use it only for trusted single-user local validation, and explicitly enable and verify authentication before exposing it to shared networks.
- Cloning the remote AI-Q repository creates external network and local filesystem effects; confirm the destination, source, and user intent before execution.
- Core operation depends on GitHub, NVIDIA model endpoints, and search providers; mainland-China reachability, proxy requirements, credentials, and regional service limits are not established.
- No commands were executed in this review; the pass claim in BENCHMARK.md does not replace deployment testing in the target environment.
What does this skill do, and when should you use it?
aiq-deploy is an NVIDIA Agent Skill for installing, deploying, running, validating, troubleshooting, and stopping NVIDIA AI-Q Blueprint infrastructure. It supports Docker Compose, local Python or CLI execution, local web UI development, and Kubernetes with Helm. After startup, it performs basic health checks and hands a reachable AI-Q server URL to aiq-research. It does not conduct deep research or assess research-report quality.
It locates or clones the NVIDIA AI-Q repository, checks required files, selects a deployment mode, creates deploy/.env only when absent, verifies ignore rules and runtime prerequisites, starts the selected services, and validates them through the /health endpoint. It checks whether required credentials are present without printing secret values; once healthy, it reports AIQ_SERVER_URL and can run optional end-to-end validation after user confirmation. It routes port, credential, configuration, compatibility, and unhealthy-service failures to the relevant troubleshooting guidance, while requiring confirmation before destructive Docker-volume cleanup.
- AI-Q users who need a local or self-hosted backend for aiq-research.
- Developers who want to start the AI-Q backend and default PostgreSQL dependency with Docker Compose.
- Teams deploying AI-Q through CLI, local web UI, Kubernetes, or Helm workflows.
- Operators who need a basic health check and a verified service URL for a downstream skill.
- Users troubleshooting port conflicts, missing credentials, configuration errors, or aiq-research compatibility problems.
What are this skill's strengths and limitations?
- Covers installation, deployment, validation, troubleshooting, shutdown, and restart workflows.
- Supports Docker Compose, local process or CLI, local web UI, Kubernetes, and Helm.
- Includes explicit safeguards for credentials, .env handling, port conflicts, and Docker-volume deletion.
- Produces a verified AIQ_SERVER_URL for aiq-research handoff.
- Requires Git, network access, and one supported runtime; some paths also require a Kubernetes cluster or GPU capacity.
- Requires NVIDIA_API_KEY and a supported search-provider key for relevant workflows; it cannot provide or inspect secret values.
- Does not perform deep research or evaluate research-report quality.
- Helm, FRAG, self-hosted models, and custom configurations depend on infrastructure controlled by the user; the source does not establish that every platform has been tested.
How do you install this skill?
Install it from the NVIDIA/skills collection with: npx skills add nvidia/skills --skill aiq-deploy --yes. Alternatively, run npx skills add nvidia/skills --list and select aiq-deploy. The README says the CLI prompts for the skill and installation destination but does not document a fixed install directory.
How do you use this skill?
Prompt a compatible Agent Skills client with a request such as “install and deploy NVIDIA AI-Q Blueprint.” If no mode is specified, the skill asks you to choose Skill backend, CLI, UI, or Custom. After deployment, the default backend uses AIQ_SERVER_URL=http://localhost:8000; for another port, set export AIQ_SERVER_URL="http://localhost:<PORT>". Configure credentials in deploy/.env or environment variables, and never paste secrets into chat.
How does this skill compare with similar options?
Choose the deployment path according to the environment: Docker Compose is the default durable local deployment; Python with uv supports local process or CLI mode; Node.js with npm supports local web UI development; kubectl with Helm supports Kubernetes deployment.