TL;DR
NVIDIA AI Blueprint VSS is a collection of reference architectures for building vision agents by integrating GPU-accelerated vision microservices, VLM, and LLM, enabling video search, summarization, and Q&A.
Key features
Real-time video intelligence: publishes feature extraction, embedding, and stream understanding results to a message broker
Downstream analysis: transforms metadata into trajectories, events, and verified alerts
Agentic and offline processing: orchestration tools (MCP support) for search, Q&A, summarization, and clip retrieval
Various workflows: Q&A and report generation, alert verification, real-time alerts, video search, long video summarization
Based on NVIDIA NIM microservices (Cosmos-Reason2-8B, Nemotron-Nano-9B-v2, etc.)
When to use it
When you need natural language search, Q&A, and summarization over large-scale video data
When rapidly prototyping or deploying vision AI agents for warehouse automation, smart space monitoring, SOP verification, etc.
For developers and researchers building real-time video analysis and agentic workflows in GPU-accelerated environments