Breaking

NVIDIA Adds Prompt-Based Agent Builder and Adaptive Sampling to VSS 3.3

October 1, 2026 at 11:59 EDT

AI Agents
Software Dev & Coding
Sales & Business

NVIDIA has released VSS Blueprint 3.3, adding a Build Vision Agent skill that composes video analytics agents from a natural-language prompt and Adaptive Efficient Video Sampling to cut vision-language model workload. In NVIDIA's demonstration, a bottling-line agent with alerts, search and reporting reached a live, previewable deployment in under 30 minutes.

Sep 29, 2026 - NVIDIA

Prompt-built visual AI agents

VSS Blueprint 3.3 adds a Build Vision Agent skill for composing video analytics workflows and Adaptive EVS to reduce the VLM work spent rereading unchanged video.

What changed

30 min

Under this time to a live previewable demo build

80%

Fewer VLM input tokens in a 60-minute summary test

46%

More concurrent real-time VLM streams

17%

Lower alert contextualization latency

Adaptive EVS raises stream capacity

Same GPU test: 13 concurrent streams before, 19 with Adaptive EVS.

The gain comes from spending visual tokens on changed regions, not repeated background frames.

How one prompt becomes a deployment

Describe goal

Cameras, alerts, search, reports

→

Pick profile

base, alerts, lvs or search

→

Compose delta

Reuse shared services

→

Validate stack

Review Compose and readiness

Why builders like it

A tested workflow can share ingestion, storage, search and reporting instead of creating a separate video pipeline for every feature.

What still needs work

Production teams still need site footage tests for accuracy, latency, false alerts, security controls and edge cases.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.