NVIDIA has introduced a specialized deep research skill for agent harnesses like Claude Code, Codex, and LangChain, designed to offload complex research tasks from general-purpose agents. While current harnesses excel at orchestration, code execution, and tool chaining, they often struggle with deep research requirements such as multi-document synthesis, decision briefs, and long-horizon analysis with source attribution. Implementing these capabilities traditionally requires teams to build extensive pipelines for data grounding, authentication, prompt tuning, and result evaluation. NVIDIA AI-Q addresses this by packaging the entire research workflow into an open-source blueprint that functions as a portable skill. With this integration, an agent harness can delegate a research task to a local or hosted AI-Q server and receive a structured report complete with citations. The AI-Q skill includes a configuration file explaining its usage and a helper script that manages request routing, job submission, and result retrieval. This approach allows the harness to focus on high-level orchestration while AI-Q handles the intricacies of intent classification, clarification, shallow and deep research, and evaluation. Crucially, sensitive source data can remain within the enterprise environment, ensuring compliance for regulated sectors such as healthcare, finance, and defense. The system supports secure integration with Model Context Protocol (MCP) servers, allowing AI-Q to pull from authenticated enterprise data sources without requiring a separate retrieval stack. NVIDIA NeMo Agent Toolkit facilitates this by treating MCP servers as function groups. The update documents three primary authentication patterns: connecting to unauthenticated servers via standard client function groups, using service accounts for batch or shared data access, and forwarding user bearer tokens for downstream APIs that trust the initiating user. Deployment flexibility is a core feature of AI-Q. The blueprint is available via Docker Compose and Helm charts, enabling installation on developer laptops, cloud Kubernetes clusters, or air-gapped data centers. This ensures that the AI pipeline runs where the data resides, meeting strict data sovereignty requirements. Furthermore, NVIDIA Nemotron open models can be self-hosted, allowing organizations to use frontier models for planning while routing sensitive research to on-premises models for compliance. The AI-Q pipeline is engineered specifically for research quality rather than general adaptation. It operates through four distinct stages, each independently tuned and evaluated against benchmarks such as FreshQA and DeepResearchBench. Teams can employ a hybrid model approach, combining reasoning models for synthesis with configurable routers for specific tasks. The system includes evaluation harnesses that measure quality against specific datasets and generates reports with detailed source attribution. This architecture ensures that agents can produce research grounded in enterprise data without directly accessing the underlying sources, providing auditability through OpenTelemetry traces. AI-Q is now available as an open-source blueprint and has been validated on Dell AI Factory. The Dell-NVIDIA AI-Q 2.0 Reference Architecture offers a production-ready solution for regulated industries, packaging the deployment patterns into a multi-agent research workflow. Teams can begin by installing the skill into their respective agent environments, linking it to their workspace for Claude or dropping it into the configured directory for Codex and OpenCode. Once installed, agents recognize a single deep research capability, routing queries for regulatory landscapes or internal policy synthesis directly to the AI-Q server for processing.