The VMware Tanzu division of Broadcom Inc. today is adding generative artificial intelligence to its platform-as-a-service that includes support for the new Model Context Protocol defined by Anthropic PBC and dissociating itself from Kubernetes, the orchestrator for software containers with which Tanzu has long been associated.

Introduced last November by Anthropic, MCP is a standardized way for AI models, particularly large language models, to interact with and access external data, tools and application programming interfaces in a structured manner. MCP support enables Tanzu users to build gen AI and agentic applications more quickly. Broadcom also announced the availability of the MCP Java software development kit within Spring, a popular open-source framework for building Java applications.

Broadcom said its AI-ready PaaS, which is optimized for private cloud, will give enterprises ways to build gen AI and agentic applications and deliver them into production in minutes.

“We are singularly focused on driving business velocity through application velocity,” said Purnima Padmanabhan (pictured), general manager of Broadcom’s Tanzu Division. “We want to accelerate ideas to code and code to production.”

Gartner Inc. forecasts that 33% of enterprise software applications will include agentic AI by 2028, up from fewer than 1% in 2024. But implementing agents such as enterprise virtual assistants is complicated by data fragmentation, Padmanabhan said.

“No application can be built without data, so we have built common data services into the platform – Postgres, MySQL, Redis, Valkey, RabbitMQ – that you can provision on demand for development or production,” she said. “The management and coordination of those services, and binding to credential management, are all taken care of.”

Single AI development platform

Spring AI supports common gen AI patterns such as chatbots and retrieval-augmented generation and frequently use data processing tasks such as memory support an function calls with abstracted API support for vector stores and all major vector databases. Tanzu Platform can push source code written in any language, including Python, to production, with integrated data services and a no-code machine learning model allowing for faster, more frequent iterations. Spring AI also includes integrated observability and utilities for AI model evaluation.

“There is a five-times improvement in developer productivity because developers don’t muck around with configuration files,” Padmanabhan said. “They just submit code and intent, and because you reduce configuration variation, you see a huge reduction in unplanned downtime as well as time to patch.”

Spring AI’s MCP implementation supports spans a wide range of use cases from simple file access to complex, multimodel connectivity for agent-based applications. Broadcom said the framework has been modified for use outside Spring and designated the official MCP Java software development kit by Anthropic.

Tanzu Platform will be integrated with Anthropic’s Claude LLM to enable direct proxy requests to Claude models with governance controls. This includes role-based access control, rate limiting for AI applications using the Claude model, the ability to review interactions in agentic flows and the ability to use the most efficient model possible, Broadcom said.

IaaS ‘dial tone’

The refreshed Tanzu includes automated, secure container builds, hard-coded dependencies like local models or data stores, connects to external models, services, pipelines and processes and deploys or updates applications without ticketing. Platform teams can curate developers’ models, automate token rate limits to manage costs, and continuously update applications and underlying operating systems to limit exposure to attacks.

Padmanabhan said this release also unifies what she called “a bag of parts” that Tanzu users had to assemble in the past, including a service mesh, data services, observability and multicluster management. Tanzu is also formally severing ties with Kubernetes, the orchestrator for software containers with which it has long been associated. Instead, the platform will assume the existence of what Padmanabhan called an “IaaS dial-tone,” referring to infrastructure-as-a-service.

“We are dissociating completely from Kubernetes,” she said. “It is a low-level technology like networking and storage. It is not an application platform.”

Going forward, Tanzu will take advantage of Kubernetes runtime and Kubernetes infrastructure management components embedded into VMware Cloud Foundation and cloud platforms. “If Kubernetes is available, we can do some more things with it, but as a pre-engineered solution, all we need is an IaaS dial tone,” she said. “We want to make this a true black box and remove complexity.”

Photo: SiliconANGLE