Learn how to build a video search and summarization agent with the new NVIDIA AI Blueprint.   Read the Blog

NVIDIA Metropolis

Discover this advanced collection of developer workflows and tools to build, deploy, and scale vision AI and generative AI from the edge to the cloud. It delivers exceptional scale, throughput, cost-effectiveness, and faster time to production.


View All Tools
NVIDIA Metropolis includes a hosts of SDK and developer tools

Click to expand

Explore All the Benefits

Faster Builds

Use and tune high-performance pretrained models to streamline AI training for your unique industry. Use our cloud-native modular microservices and reference applications to accelerate development.

Lower Cost

Powerful SDKs—including NVIDIA TensorRT™, DeepStream, and TAO Toolkit—reduce overall solution cost by maximizing inference throughput and optimizing hardware usage on NVIDIA platforms and infrastructure.

Flexible Deployments

Deploy with flexibility using cloud-native Metropolis microservices and containerized apps offering options for on-premises, cloud, or hybrid deployments.



Enhance Your Applications With Generative AI

Create Visual AI Agents With NVIDIA AI Blueprint

The NVIDIA AI Blueprint helps enterprises build visual agents for video search and summarization capable of understanding activity within massive volumes of live or archived video. These agents are powered by generative AI, vision language models (VLMs), large language models (LLMs), and NVIDIA NIM™ microservices. Just give them tasks through natural language prompts, and they can perform complex assignments like video summarization or visual question answering, unlocking entirely new application possibilities.


Learn More
NVIDIA Video Insight Agent (VIA) searches videos with natural language interactions on a laptop
Operational conveyor belt moving boxes in a warehouse

Retrieval Augmented Generation (RAG) Workflow

This suite of customizable, cloud-native building blocks is ideal for developing generative AI applications. The workflows let you accelerate development of generative AI applications by seamlessly integrating Large Language Models (LLM) with enterprise data.


Watch GTC Talk



Metropolis APIs and Microservices

With the rapidly evolving AI landscape, developers building vision AI applications for the edge are challenged by more complex and longer development cycles. NVIDIA Metropolis offers a collection of powerful APIs and microservices for developers to easily develop and deploy applications on the edge to any cloud.


Learn More

Powerful Tools for
AI-Enabled Video Analytics

The Metropolis suite of SDKs provides a variety of starting points for AI application development and deployment.



NVIDIA Omniverse™ Replicator

Generate physically accurate 3D synthetic data at scale, or build your own synthetic data tools and frameworks. Bootstrap perception AI model training and achieve accurate Sim2Real performance without having to manually curate and label real-world data.


Learn More About Omniverse Replicator

Also check out Omni.Replicator.Character, an Isaac Sim extension using Replicator that focuses on simulation and synthetic data generation (SDG) of agent movement—such as from people and an Autonomous Mobile Robot (AMR)—using a high-level user interface.


Learn More About Omniverse Replicator
An autonomous mobile robot in action in a warehouse
Choose from 100+ model architectures and 25k+ task-based AI models

Pretrained Models

Eliminate the time-consuming process of building models from scratch. Choose from over 100+ permutations of highly accurate models and generic neural network architectures or start with our task-based models to recognize human actions and poses, detect people in crowded spaces, classify vehicles and license plates, and much more.


Learn More About Pretrained AI Models Try Pretrained Models With Jupyter Notebook

TAO Toolkit

The Train, Adapt, and Optimize (TAO) Toolkit is a low-code AI model development solution that lets you use the power of transfer learning to fine-tune NVIDIA pretrained models with your own data and optimize for inference—without AI expertise or a large training dataset.


Learn More About TAO Toolkit Try TAO on LaunchPad
Leverage NVIDIA TAO Toolkit to train, adapt, and optimize AI model development
Logos of popular AI model frameworks

Many AI Model Frameworks

Create your AI models and applications on these popular NVIDIA-supported AI frameworks. Integrate any existing AI model into the Metropolis workflow and easily customize existing models in TensorFlow, PyTorch, and more by easily converting to TAO.


Learn More About Deep Learning Frameworks

TensorRT

This SDK for high-performance deep learning inference includes an inference optimizer and runtime that delivers low latency and high throughput, both on edge devices and within the cloud. TensorRT is supported on all popular frameworks, including TensorFlow and PyTorch. Powering NVIDIA solutions such as NVIDIA JetPack™ and DeepStream, TensorRT give you a gateway to accelerated inferencing.


Learn More About TensorRT
TensorRT SDK for high-performance deep learning inference
NVIDIA DeepStream SDK - a complete streaming analytics toolkit

DeepStream SDK

NVIDIA DeepStream SDK is a complete streaming analytics toolkit based on GStreamer for AI-based multi-sensor processing, video, audio, and image understanding. It’s ideal for vision AI developers, software partners, startups, and OEMs building IVA apps and services.


Learn More About DeepStream SDK Try DeepStream on LaunchPad

Triton Inference Server

The NVIDIA Triton™ open-source, multi-framework inference serves software to deploy, run, and scale AI models in production on both GPUs and CPUs. It supports all major frameworks, including TensorFlow and Pytorch, and maximizes inference throughput on any platform.


Learn More About Triton Inference Server
Triton Inference Server maximizes inference throughput on any platform
Video Storage Toolkit (VST) manages and stores footage for large volumes of video cameras

Video Storage Toolkit (VST)

Easily manage and store footage for large volumes of video cameras with hardware-accelerated video decoding, streaming, and storage. Get started quickly with the included web-based user interface and take advantage of VST flexibility through intuitive REST APIs. It’s available for NVIDIA Jetson Xavier™ and Orin™ devices.


Learn More About VST

Metropolis Microservices

This suite of cloud-native microservices and reference applications fast-tracks development and deployment of vision AI applications. Unlock business insights for a wide range of spaces, ranging from roadways to airports to retail stores, in significantly shortened development cycles.


Learn More About Metropolis microservices
Use Metropolis Microservices to develop and deploy vision AI apps
NVIDIA CUDA-X libraries help with pre-processing and model performance

CUDA-X Libraries

Take advantage of low-level libraries and primitives for computer vision and more that can help with pre-processing and model performance. NVIDIA® CUDA-X™, built on top of NVIDIA CUDA®, is a collection of libraries, tools, and technologies that deliver dramatically higher performance in compute-intensive algorithms spanning complex math, deep learning, and image processing.


Cloud Containers

Containerize your applications easily with Docker, Kubernetes, and GPU Operators to deploy cloud-native solutions on Jetson, x86, and dGPU. Or use pre-built cloud-native Metropolis microservices.


Learn More About Cloud Containers
Video Storage Toolkit (VST) manages and stores footage for large volumes of video cameras

Generate – Synthetic Data Generation

NVIDIA Omniverse™ Replicator

Generate physically accurate 3D synthetic data at scale, or build your own synthetic data tools and frameworks. Bootstrap perception AI model training and achieve accurate Sim2Real performance without having to manually curate and label real-world data.


Learn More About Omniverse Replicator

Also check out Omni.Replicator.Character, an Isaac Sim extension using Replicator that focuses on simulation and synthetic data generation (SDG) of agent movement—such as from people and an Autonomous Mobile Robot (AMR)—using a high-level user interface.


Learn More About Omniverse Replicator
An autonomous mobile robot in action in a warehouse

Train – Application-Specific Model Customization

Pretrained AI Models

Pretrained models

Eliminate the time-consuming process of building models from scratch. Choose from over 100+ permutations of highly accurate models and generic neural network architectures or start with our task-based models to recognize human actions and poses, detect people in crowded spaces, classify vehicles and license plates, and much more.


Learn More About Pretrained AI Models Try Pretrained Models With Jupyter Notebook

TAO Toolkit

The Train, Adapt and Optimize (TAO) Toolkit is a low-code AI model development solution that lets you use the power of transfer learning to fine-tune NVIDIA pretrained models with your own data and optimize for inference–without AI expertise or a large training dataset.


Learn More About TAO Toolkit Try TAO on LaunchPad
Leverage NVIDIA TAO Toolkit to train, adapt, and optimize AI model development
Logos of popular AI model frameworks

Many AI model frameworks

Create your AI models and applications on these popular NVIDIA-supported AI frameworks. Integrate any existing AI model into the Metropolis workflow and easily customize existing models in TensorFlow, PyTorch, and more by easily converting to TAO.


Learn More About Deep Learning Frameworks

Build – Powerful AI applications

TensorRT

This SDK for high-performance deep learning inference includes an inference optimizer and runtime that delivers low latency and high throughput, both on edge devices and within the cloud. TensorRT is supported on all popular frameworks, including TensorFlow and PyTorch. Powering NVIDIA solutions such as JetPack™ and DeepStream, TensorRT is a gateway to accelerated inferencing.


Learn More About TensorRT
TensorRT SDK for high-performance deep learning inference
NVIDIA DeepStream SDK - a complete streaming analytics toolkit

DeepStream SDK

NVIDIA DeepStream SDK is a complete streaming analytics toolkit based on GStreamer for AI-based multi-sensor processing, video, audio, and image understanding. It’s ideal for vision AI developers, software partners, startups, and OEMs building IVA apps and services.


Learn More About DeepStream SDK Try DeepStream on LaunchPad

Triton Inference Server

The NVIDIA Triton™ open-source, multi-framework inference serves software to deploy, run, and scale AI models in production on both GPUs and CPUs. It supports all major frameworks, including TensorFlow and Pytorch, and maximizes inference throughput on any platform.


Learn More About Triton Inference Server
Triton Inference Server maximizes inference throughput on any platform
Video Storage Toolkit (VST) manages and stores footage for large volumes of video cameras

Video Storage Toolkit (VST)

Easily manage and store footage for large volumes of video cameras with hardware-accelerated video decoding, streaming, and storage. Get started quickly with the included web-based user interface and take advantage of VST flexibility through intuitive REST APIs. It’s available for NVIDIA Jetson Xavier™ and Orin™ devices.


Learn More About VST

Metropolis Microservices

This suite of cloud-native microservices and reference applications fast-tracks development and deployment of vision AI applications. Unlock business insights for a wide range of spaces, ranging from roadways to airports to retail stores, in significantly shortened development cycles.


Learn More About Metropolis microservices
Use Metropolis Microservices to develop and deploy vision AI apps
 NVIDIA CUDA-X libraries help with pre-processing and model performance

CUDA-X libraries

Take advantage of low-level libraries and primitives for computer vision and more that can help with pre-processing and model performance. NVIDIA® CUDA-X™, built on top of NVIDIA CUDA®, is a collection of libraries, tools, and technologies that deliver dramatically higher performance in compute-intensive algorithms spanning complex math, deep learning, and image processing.


Deploy – Application Management and Scaling

Video Storage Toolkit (VST) manages and stores footage for large volumes of video cameras

Cloud containers

Containerize your applications easily with Docker, Kubernetes, and GPU Operators to deploy cloud-native solutions on Jetson, x86, and dGPU. Or use pre-built cloud-native Metropolis microservices.


Learn More About Cloud Containers

Get Started With Sample Applications

An engineer deploys NanoOwl object detection and metadata output on Redis in a research room

Generative AI at the Edge

Learn how to bring generative AI applications into production faster with Metropolis microservices for Jetson.

Read the Blog
Integrate TAO with DeepStream for face mask detection

Vision AI Applications With APIs

Learn how to build vision AI applications using APIs provided by Metropolis microservices for Jetson.

Read the Blog
Use TAO toolkit to detect bottle defect in a factory

Defect Detection

Learn how to transform industrial defect detection with NVIDIA TAO and vision AI models.

Read the Blog
Use TAO with Deepstream for number plate detection

Character Detection and Recognition

Learn how to train and deploy character detection and recognition models with NVIDIA TAO and Triton™.

Read the Blog

View all Metropolis technical blogs

Explore NVIDIA GTC Talks On-Demand

Develop, deploy, and scale AI-enabled video analytics applications with NVIDIA Metropolis.


Get Started