From Experimental Agents to a Centralized AI Command Center
Enterprises are rapidly moving from single chat assistants to fleets of autonomous agents embedded in finance, supply chain and manufacturing systems. That shift is creating a new governance problem: AI assets are multiplying across business units and vendors with little visibility or control. At the Sapphire conference, SAP responded with the SAP AI Agent Hub, positioned as a vendor-agnostic command center for AI agent governance. Initially tied to SAP LeanIX, the hub is now being opened up through Joule Studio so more customers can consolidate oversight of every AI agent, large language model (LLM) and Model Context Protocol (MCP) server they run. The goal is not just better inventory, but a foundation for enterprise LLM management that aligns AI behavior with existing process, identity and compliance controls, before agents touch systems of record or operate autonomously at scale.
Taming Vendor Agent Sprawl with a Unified Registry
Most large organizations are on track to run hundreds of agents from multiple vendors: productivity copilots, CRM agents, AI-native tools and custom-built workflows. Today, those agents typically live in siloed systems with no central inventory or audit trail for IT and security teams. SAP’s AI Agent Hub targets this vendor agent sprawl by offering a single system of record that automatically discovers agents, LLMs and MCP servers across the enterprise, regardless of where or how they were built. The hub’s AI registry, now generally available, is the cornerstone of this AI agent management platform. It collects authoritative metadata on each asset and then layers on governance capabilities: workflow evaluation, risk ratings and compliance mappings ensure agents cannot be promoted to production without a verified governance record. In effect, the registry turns a chaotic agent landscape into a manageable portfolio with traceable ownership and lifecycle state.
Identity, Observability and Process-Aware ‘Agent Mining’
Beyond inventory, SAP is building a control fabric around agents. Planned capabilities include identity and access control via SAP Cloud Identity Services, giving each agent a unique identity by default. That enables fine-grained authorization, data access policies and auditable interactions between agents and core business systems. AI observability, also targeted for release later, will add session-level telemetry on tool calls, health and root-cause analysis—similar to what specialized observability vendors provide today. This lets teams understand who is interacting with which agents, whether tools are used efficiently and when human-in-the-loop steps are still required. SAP Signavio extends the concept further with “agent mining,” applying process mining techniques to compare designed execution pathways against how non-deterministic agents actually behave. That process-aware view allows enterprises to detect drift, inefficiencies and compliance risks in live agent workflows, closing the loop from design to runtime behavior.
NVIDIA Collaboration: Secure Runtime for Production-Grade Agents
The AI Agent Hub sits within SAP’s broader Business AI Platform, which now embeds NVIDIA OpenShell as the runtime security layer for all SAP AI agents, including custom Joule Studio agents. OpenShell provides isolated execution environments, policy enforcement at filesystem and network layers, and infrastructure-level containment to limit damage when agent logic fails. This hardens the runtime so enterprises can ask, at a technical level, whether a particular agent action can safely execute. On top of that, Joule Studio’s control layer focuses on whether the action should happen at all, based on business policies, roles and processes. By co-developing OpenShell, SAP and NVIDIA are aligning runtime hardening, policy modeling, identity integration and auditing to the needs of enterprise LLM management. Together with the AI Agent Hub, this stack aims to make autonomous agents trustworthy enough for production-scale business operations without locking customers into a single AI vendor.
Why SAP’s Installed Base Gives It an AI Governance Edge
SAP’s strategic advantage in AI agent governance is its existing footprint across enterprise architecture, processes and identities. LeanIX already maps IT landscapes, Signavio documents actual business workflows, Cloud Identity Services manage users and roles, and SuccessFactors maintains organizational structures. The AI Agent Hub can tap into this context to ground AI agent governance in how the business really runs, rather than treating agents as isolated technical components. As organizations adopt agents from Microsoft, Salesforce, OpenAI, Anthropic and open frameworks, SAP is positioning the hub as the neutral fabric that inventories, secures and monitors them all. By remaining vendor-agnostic at the agent, LLM and MCP levels while tightening integration with its own platforms, SAP aims to solve vendor agent sprawl without enforcing a monolithic stack. For CIOs and CISOs, that combination of broad coverage and strong guardrails could be the difference between AI-driven efficiency and unmanageable AI risk.