A private cloud platform for AI agents — and the apps that run alongside them.

What's new in 10.4

The PaaS, extended for agents.

Tanzu Platform has been a PaaS for over a decade - Cloud Foundry roots, Java and Spring integration to the bones, the day-two stuff most teams forget about until they're running it in production. The 10.4 release extends that to the new shape of work: AI agents, MCP servers, and the harnesses being built around them.

For agentic apps

Agent Foundations.

• Deny-by-default runtime Zero-trust networking, structural secrets isolation, and an isolated agent sandbox so agents can't wander where they shouldn't.
• MCP Gateway Every tool call routed through one place. OIDC "digital passports" tie each agent's actions back to a verifiable identity, with an audit trail you can actually use.
• Service marketplace A curated catalog of MCP servers, LLMs, databases, and ISV tools. Self-service binding, with credentials injected by the platform - developers and agents never handle secrets.
• Persistent agent memory Long-term memory via Tanzu Postgres with pgvector and Spring AI chat memory. Useful for the agent-as-coworker scenarios that span hours or days.
• Showback & chargeback Per-application cost visibility by service, space, or org. So you can find which agent is burning through the budget.
• Agent buildpack (tech preview) Curated execution framework for Java/Spring AI and Python. Bind a model, integrate MCP servers and private data, deploy.

For everything else

Still a PaaS.

It's easy to lose track in an AI announcement, but Tanzu Platform still does what it's always done: a private cloud platform for running applications, integrated with Spring, self-service services and data, and the platform engineering bits that make day-two manageable. Over a decade running some of the most critical applications out there.

Build apps in Java, .NET, Node.js, Python, Go, Ruby, PHP, static sites - and any other binary or Docker image you want to deploy.

The data side

Because agents are only as smart as the data they reach.

Tanzu Data Intelligence 10.4 ships alongside: a SQL Assistant for natural-language queries against Greenplum, a Greenplum MCP server for AI-assisted admin, Apache Iceberg support (tech preview), and selective DR filtering. Read-only guardrails on the LLM, so it can't DROP your tables.

Start

Ninety days. Real platform. Real apps.

Read the kit, stand up Tanzu Platform for ninety days, run a real PoC. If it doesn't earn its keep, you've spent a quarter tire-kicking - which is what you were going to do anyway trying to build your own platform. At least you'll be ready on day 91.

The official stuff

Prefer the official data sheet and snazy web pages experience? See the Tanzu Platform overview, the docs, and demos and videos. Or read Darin Zook's 10.4 launch post and Camille Crowell-Lee's deep dive on Agent Foundations.