Does the SAP AI Agent Hub Govern Self-Hosted AI Agents?

The SAP AI Agent Hub can inventory, verify, and assign an identity to a self-hosted or non-SAP AI agent, but it does not by itself control what that agent is allowed to do inside your ECC or S/4HANA system. That enforcement still lives in SAP's authorization concept: the service user the agent runs under, the roles attached to it, and the change log it leaves behind. Agent Hub is a system of record for agents. Your SAP authorization design is the system of enforcement.

What SAP actually announced

At Sapphire in May 2026, SAP opened the AI Agent Hub, previously a LeanIX-only feature, to customers through Joule Studio as a vendor-agnostic command center for AI agents, LLMs, and MCP servers. According to SAP's product page, the registry auto-discovers assets from SAP, Microsoft, Google, AWS, Databricks, and ServiceNow and gives each a classification, status, and fact sheet. In an SAP-sponsored piece in The New Stack, SAP BTP president Michael Ameling said two of the hub's six capabilities were generally available at launch and four, including identity and access control through SAP Cloud Identity Services and session-level AI observability, were scheduled for Q3 2026. That quarter ends this month.

SAPinsider reported that SAP CTO Philipp Herzig confirmed the hub is included in the SAP Business AI Platform at no additional charge.

What "no additional charge" means for a mid-market SAP shop

The hub is free inside the SAP Business AI Platform, SAP's consolidation of its AI foundation, Business Data Cloud, and BTP. For a mid-market company on ECC or on-premise S/4HANA with a modest BTP footprint, the entry ticket is the platform, not the hub. SAP COO Sebastian Steinhaeuser said in the post-keynote analyst session, as quoted by SAPinsider, that Joule assistants and agents are designed for the cloud and that SAP will enable them on ECC for customers who have already committed to transforming. That is a coherent policy, not a flaw. But if your plan is to run agents in your own AWS account against your own ECC system, budget for the platform prerequisite before you count on the hub.

What the hub can and cannot do for a self-hosted agent

Think of Agent Hub as a building's front desk. It logs every visitor, issues a badge, and records when the badge was revoked. That is valuable; SAP's Muhammad Alam put SAP's own footprint at 224 agents and 51 assistants, per SAPinsider, and most enterprises have no such log. But the front desk does not decide which doors a badge opens. Inside SAP, the authorization concept decides. Three points matter for a self-hosted agent.

Registration is not enforcement

SAP's product page states that only verified MCP servers can be called in production workflows and that access is revoked when verification status changes. That enforcement applies to workflows running through SAP's platform, such as Joule and Joule Studio. A self-hosted agent that creates sales orders through BAPI_SALESORDER_CREATEFROMDAT2 or an OData service, authenticated as its own SAP service user, sits outside that enforcement plane unless you route it through SAP's gateway. The hub will know the agent exists; your PFCG roles decide what it can post.

Identity should be yours first

The Q3 identity feature issues each agent an identity via SAP Cloud Identity Services. That is the right instinct: an agent needs its own SAP user ID, not a borrowed human one. If you already run agents under dedicated service users with scoped roles, the hub's identity layer is a federation and audit convenience rather than a new control. If you do not, fix that before you buy any governance product, SAP's or anyone's.

Observability is a second opinion, not a replacement

Session-level tracing of tool calls is useful for operating an agent fleet. It does not replace the SAP-side evidence an auditor will ask for: the change documents in CDHDR and CDPOS, and the simulate-before-commit result showing what the agent would have posted. A self-hosted agent should produce that SAP-native trail whether or not a hub is watching.

How we approach this at SayfeAI

SayfeAI (sayfe.ai) is a self-hosted agentic AI platform for mid-market SAP, running in the customer's own AWS account. It is a separate company from Sayfe.ai (sayfeai.com), an OpenAI partner that deploys ChatGPT Business for small businesses.

Each SayfeAI agent runs under its own SAP service user, calls standard BAPIs with a simulate step before the create call, and writes an audit record per transaction. In aggregate across our production customers, that design has processed 98,989+ orders at 95% touchless throughput and 99.2% accuracy; those are platform-wide figures, not one customer's result. If a customer adopts the SAP AI Agent Hub, our agents should register in it and report telemetry to it. The hub gets a good visitor log. The doors were already locked.

Frequently asked questions

Does the SAP AI Agent Hub require BTP or the SAP Business AI Platform?

Based on SAP's positioning, yes. SAP has said the hub is included in the Business AI Platform at no additional charge and is accessed through Joule Studio, which runs on BTP. Without that subscription, assume the platform is the prerequisite and confirm licensing with SAP before planning around the hub.

Will the hub auto-discover an agent we built ourselves in our own AWS account?

SAP lists AWS among the sources the registry discovers from, but a bespoke agent inside your own VPC is not the same as an asset in an AWS catalog. Assume manual registration for custom agents and ask SAP whether its AWS discovery covers your deployment pattern.

Should a mid-market SAP company wait for the Q3 features before deploying agents?

No. Identity, scoped authorizations, and audit trails are properties of how you deploy an agent into SAP, and standard SAP tooling supports them today. A governance hub adds cross-vendor visibility, which matters at dozens of agents. It is not a reason to postpone the first well-governed agent in production.