Yes, NVIDIA Agent Toolkit deploys on all major cloud providers: Amazon Web Services (AWS), Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure. The toolkit and OpenShell can be accessed via build.nvidia.com and deployed on NVIDIA Cloud Partners including Baseten, CoreWeave, DeepInfra, Fireworks AI, Lambda, Lightning AI, Together AI, and Vultr. Each cloud provider path supports GPU acceleration through NVIDIA GPUs available on their platforms.
For cloud deployment, containerization is standard—package your agent code and NVIDIA toolkit dependencies in Docker, then deploy to Kubernetes on EKS (AWS), GKE (Google Cloud), AKS (Azure), or OKE (Oracle). NVIDIA provides Helm charts and deployment guides for major cloud platforms. Managed inference services like Baseten and CoreWeave handle scaling, monitoring, and failover automatically. Your agents run as persistent services accessible via REST or gRPC APIs.
Cloud deployment pairs naturally with fully-managed vector databases. You can run Milvus in cloud-managed Kubernetes, use Zilliz Cloud (the managed Milvus service), or integrate with cloud-native vector services. This enables elastic scaling: as agent traffic increases, the cloud platform automatically scales both the agent runtime and retrieval layer.
For workloads requiring on-premises data residency, agents can run on-cloud while vector databases remain on-premises, connected securely via private networking. This hybrid approach satisfies compliance requirements while leveraging cloud scalability. Agent memory layers benefit significantly from vector database integration. Milvus provides semantic search across stored embeddings, allowing agents to retrieve contextually relevant information for reasoning tasks. Learn more about retrieval-augmented generation with Milvus and explore Zilliz Cloud for enterprise deployments.