Give Codex a Deep Research Skill With NVIDIA AI-Q

Developers building agent workflows often need their agents to research across multiple sources, synthesize findings, and return answers with citations. But, rebuilding that deep research pipeline inside every harness adds extra work. NVIDIA AI-Q gives agent harnesses like Codex a reusable way to delegate deep research instead of recreating retrieval, reasoning, synthesis, and citation logic in each agent. In this demo, Codex calls the AI-Q research skill, sends a multi-source research request to a running AI-Q server, follows the async checkpoint flow, and receives a structured report with citations. The walkthrough shows the basic architecture: Codex runs on the developer workstation, AI-Q handles the research workflow on a local or hosted server, and hosted NVIDIA inference models support planning, research, and embeddings. Use this pattern with Codex, Claude Code, OpenCode, or other agent harnesses to add grounded, source-attributed research to existing workflows. ➡️ Learn more: https://github.com/NVIDIA-AI-Blueprints/aiq#getting-started 📝 Tech blog: https://developer.nvidia.com/blog/add-a-specialized-deep-research-skill-to-agent-harnesses/ 📥 Download: https://github.com/NVIDIA-AI-Blueprints/aiq/tree/develop/.agents/skills/aiq-research 00:00 - Architecture Overview 00:30 - Install the AI-Q Research Skill 01:15 - Submit a Deep Research Query from Codex 02:05 - AI-Q Runs the Async Research Workflow 03:15 - Review the Structured Report and Citations 04:05 - Execution Trace and Skill Latency