NVIDIA AI-Q Blueprint

NVIDIA AI-Q Blueprint is an enterprise-grade research agent from NVIDIA's official AI Blueprint series, built on NeMo Agent Toolkit and LangGraph state machine architecture. The system uses an Orchestration Node for intent classification and automatic routing: simple queries take the shallow research path (bounded, fast, tool-augmented), while complex queries enter the deep research path (multi-step planning, iterative retrieval, long-form report generation), all with traceable citations.

Workflows are driven by YAML configuration files, enabling zero-code adjustment of Agent combinations, tool selection, and model pairing. The default model combination uses Nemotron 3 Nano 30B + GPT-OSS 120B, with options to switch to frontier models like GPT-5.2. Data sources support Tavily Web search, Google Scholar academic search, Exa, and LlamaIndex knowledge layer or external Foundational RAG services.

The system includes built-in evaluation benchmarks — Deep Research Bench (RACE + FACT evaluation) and FreshQA (temporal fact evaluation) — for continuously measuring Agent output quality. Multiple run and deployment options are available: CLI, Web UI (FastAPI + React), Docker Compose, and Helm Chart (Kubernetes), with optional Phoenix Tracing integration for execution traceability. The project uses a uv workspace monorepo structure, registering Agent components via nat.plugins entry-points.

Agent Topology:

Orchestration Node
├── Meta Response (greetings, capability descriptions)
├── Shallow Researcher (bounded, fast, tool-augmented)
├── Deep Researcher (multi-stage: planning → iterative retrieval → citation management → report generation)
└── Clarifier

Environment Requirements: Python ≥3.11, <3.14; Node.js 22+ (Web UI only); package manager uv (recommended) or pip.

Key Environment Variables:

• NVIDIA_API_KEY (required): NVIDIA NIM API inference
• TAVILY_API_KEY (optional): Tavily Web search
• SERPER_API_KEY (optional): Google Scholar academic search
• OPENAI_API_KEY (optional): only for config_frontier_models.yml

Core Configuration Files:

• config_cli_default.yml: CLI default, Nemotron 3 Nano 30B + GPT-OSS 120B
• config_web_default_llamaindex.yml: Web default, with LlamaIndex knowledge retrieval
• config_web_frag.yml: Helm default, with Foundational RAG
• config_frontier_models.yml: Mixed frontier models, GPT-5.2 for orchestration/planning