aiq-deploy
Use when asked to install, deploy, run, validate, troubleshoot, or stop NVIDIA AI-Q Blueprint infrastructure.
Security Assessment
Detected risks:
About aiq-deploy
aiq-deploy is an operations skill for getting a local or self-hosted NVIDIA AI-Q Blueprint server installed, deployed, validated, troubleshooted, and stopped. It owns the full lifecycle of standing up the AI-Q backend so that a companion skill, aiq-research, can run deep research against a verified server URL. The skill deliberately does not perform research itself: its success criterion is a deployed and basically validated server, after which it hands off the confirmed AIQ_SERVER_URL. The workflow is kept explicit so deployment validation and handoff stay repeatable across supported agent clients.
The skill follows a numbered workflow: locate or clone the AI-Q repository (https://github.com/NVIDIA-AI-Blueprints/aiq), confirm expected files (pyproject.toml, deploy/.env.example, configs), select a deployment mode, prepare deploy/.env from deploy/.env.example without overwriting secrets, check prerequisites, start services, run validation, and report the URL. It supports four deployment paths: Skill backend (backend-only for aiq-research), CLI (interactive terminal), UI (browser app with backend and frontend on port 3000), and Custom. Docker Compose mode starts the AI-Q backend and PostgreSQL by default. Validation includes a health check via curl -sf http://localhost:8000/health, and the default URL is AIQ_SERVER_URL=http://localhost:8000. A routing table maps user intents to reference docs covering env-and-secrets, configs, docker-compose, kubernetes-helm, FRAG, troubleshooting, shutdown, and end-to-end validation.
It targets operators and developers deploying AI-Q on Claude Code, OpenCode, Codex, and other Agent Skills-compatible tools. Requirements include Git, GitHub network access, and one runtime path: Docker Compose v2, Python 3.11+ with uv, Node.js 20+ with npm, or kubectl 1.28+ with Helm 3.12+. Hosted-model usage needs NVIDIA_API_KEY, and web research needs a search provider key such as TAVILY_API_KEY, SERPER_API_KEY, or EXA_API_KEY. Credentials must stay outside chat, and deploy/.env should be git-ignored (checked with git check-ignore). The skill is designed for Blueprint version 2.1.0 and follows semantic versioning compatibility rules where major versions must match.
FAQ
What deployment modes does aiq-deploy support?
Four paths: Skill backend (backend-only service for aiq-research without a browser UI), CLI (interactive terminal AI-Q), UI (browser app with backend and frontend), and Custom (choose an existing config or review advanced customization docs). Docker Compose v2 is the default durable local deployment and starts the backend plus PostgreSQL.
What are the prerequisites?
Git, network access to GitHub and NVIDIA-hosted model endpoints, and one runtime: Docker Compose v2, Python 3.11+ with uv, Node.js 20+ with npm, or kubectl 1.28+ and Helm 3.12+. Hosted-model use requires NVIDIA_API_KEY, and web research requires a search provider key such as TAVILY_API_KEY, SERPER_API_KEY, or EXA_API_KEY.
How are secrets handled?
Secrets are stored in deploy/.env, created from deploy/.env.example without overwriting existing values. Credentials must be kept outside chat and never printed. Before writing secrets, verify deploy/.env is git-ignored using git check-ignore deploy/.env, and fix the ignore rule if it is not.
How does it validate the deployment and hand off?
After startup it runs mode-appropriate checks; for the default local backend it verifies health with curl -sf http://localhost:8000/health. It then reports the verified AIQ_SERVER_URL (default http://localhost:8000) for aiq-research, or tells the user to export a custom URL if the backend runs elsewhere.
Which Blueprint version is this skill compatible with?
It is designed for NVIDIA AI-Q Blueprint version 2.1.0. Compatibility requires the major version to match and the Blueprint minor version to be equal or greater; patch versions do not affect compatibility. For example, 2.1.0 works with 2.1.5 and 2.2.0 but not 3.0.0 or 2.0.0.
All Files
18 filesInstall aiq-deploy
Quick Setup:
- Copy the skill folder to
.claude/skills/ - Claude will automatically detect and use the skill
Repository
nvidia/skills