Agentic AI Platform on STACKIT with Kubernetes

Motivation

For most organizations, the challenge with generative and agentic AI is no longer the prototyping stage, but rather operating the technology reliably on a large scale while retaining data sovereignty. Credentials are duplicated across teams, model consumption cannot be attributed to its source, each agent integrates their own tools, and deployment practices diverge from project to project. Where regulation or sovereignty requirements apply, routing sensitive prompts through external LLM APIs raises further concerns.

This offering establishes a governed platform built on STACKIT as a sovereign cloud, ensuring that data and AI workloads remain under German jurisdiction. A central AI gateway provides a single point of access to all models (STACKIT managed or self-hosted) and to the governed MCP tool servers. This means that access control, budgets and guardrails can be enforced centrally rather than on a project-by-project basis.

What we bring

With more than two decades of experience in enterprise software development, cloud-native architecture, and Kubernetes operations, PRODYNA is well-placed to support your business. We deliver a governed, scalable Agentic AI platform built on standard, enterprise-grade open-source technologies.

• A central open-source AI gateway (AgentGateway) for unified, OpenAI compatible access to STACKIT-managed and self-hosted models with single sign-on, RBAC at group level, credential injection, token budgets, guardrails, and cost attribution at agent level
• A governed MCP gateway that fronts Model Context Protocol tool servers with shared authentication, per-server authorisation, and a self-service registry so that agents only discover the tools they are permitted to use
• A ready-to-use chat UI (LibreChat) with single sign-on, per-user model menus, token budgets and MCP tools in chat
• A GitOps-based delivery workflow (CI/CD, hardened container images and ArgoCD) for zero-downtime deployments
• Built-in observability with the Grafana stack: Prometheus metrics, Grafana Tempo (optional Langfuse) for distributed LLM & agent tracing, and ready-made dashboards for usage, cost and budgets
• Optional self-hosted LLMs on Kubernetes (vLLM) can be managed optionally by KServe
• Hands-on enablement: we develop the initial AI agents alongside your team

What you need

To make the best use of this offer and permit a fast and efficient start, you will need:

• A STACKIT Organization + Foundation (Landing Zones)
• Availability of your experts (e.g. platform, network, IAM, security)
• A small group of motivated IT colleagues to onboard as early adopters

What you get