---
title: How do you govern agents built in Joule Studio on SAP Build?
url: https://opencontroller.lyzr.com/blog/govern-sap-joule-studio-agents
published: 2026-11-17
updated: 2026-11-17
author: Lyzr
---

# How do you govern agents built in Joule Studio on SAP Build?

**Govern Joule Studio agents at five points: who may build them, which tools and systems they can reach, what data grounds them, how a new version is reviewed before release, and how they appear in your estate-wide inventory. SAP supplies identity propagation and runtime tracing on SAP AI Core. Your job is the decisions around them.**

Joule is SAP's assistant, and the agent builder in Joule Studio lets teams build their own Joule agents in SAP Build. SAP announced general availability of the agent builder around the turn of 2026. The broader question of how Joule respects SAP roles is covered in [governing SAP Joule agents](/blog/govern-sap-joule-agents). This guide focuses on custom agents your teams build.

## What does SAP provide for Joule Studio agents?

These points come from SAP's own architecture guidance and product material. Check SAP Help Portal for your release.

| Area | What SAP describes |
|---|---|
| Runtime | Agents run on SAP AI Core with built-in metering, tracing and security |
| Publishing | Deployed agents become available in Joule and are registered in its catalog |
| Tools | SAP and third-party systems via REST and OData APIs; SAP Build Process Automation workflows and business rules; MCP connections configured in SAP BTP |
| Grounding | Curated document sets and operational data through SAP AI Core |
| Identity | Inherits SAP Cloud Identity Services and principal propagation, so actions stay within the user's authorisation |
| Oversight | Human approval by default for critical decisions; data protection guardrails |
| Pro-code path | SAP Cloud SDK for AI and frameworks such as LangGraph, connected to Joule over A2A |

The identity model is the strongest built-in control. Because agents act within the user's authorisation, an agent cannot do what its user could not. That also means an over-privileged user makes an over-privileged agent.

## Which controls should you add?

1. **Restrict who can build and deploy.** Limit Joule Studio access to named builders and separate the right to build from the right to deploy to production.
2. **Approve each tool.** Every API, workflow and MCP connection an agent uses should be on an approved list with an owner. Treat a new MCP connection like a new integration. See [governing MCP servers](/blog/how-to-govern-mcp-servers).
3. **Classify grounding data.** Record which document sets each agent is grounded on and their sensitivity. Remove documents that users of the agent should not see.
4. **Review before release.** Test new versions against a fixed set of prompts, including misuse cases, and record the result. Keep the approval with the version.
5. **Use the traces.** SAP AI Core tracing tells you what happened. Decide who reviews it and how often, and how long it is kept.
6. **Clean up roles first.** Because agents inherit the user's authorisations, remove segregation-of-duties conflicts and dormant high-risk roles before wide rollout.
7. **Isolate workloads.** Use separate SAP AI Core resource groups for teams or risk tiers. See [SAP AI Core resource groups](/blog/sap-ai-core-resource-groups-for-agent-isolation).
8. **Watch model access.** If agents or pro-code variants use the generative AI hub, apply its masking and content filtering. See [generative AI hub governance](/blog/sap-generative-ai-hub-governance).

## What does Joule Studio not cover by itself?

The Joule catalog lists Joule agents. It does not list the agents your company runs on Amazon Bedrock AgentCore, Microsoft Foundry, Salesforce or developer laptops. Agents often call each other across those boundaries. An estate-wide registry, owners, and consistent policies sit above any single platform. This is the core idea in [governing AI agents across platforms](/blog/how-to-govern-ai-agents-across-platforms).

## What should a Joule Studio agent record contain?

| Field | Example |
|---|---|
| Agent name and version | Supplier onboarding assistant, v3 |
| Business owner and builder | Procurement operations; named builder |
| Tools and MCP connections | Supplier API, approval workflow |
| Grounding sources | Supplier policy documents |
| Users and roles | Procurement clerks with named role collections |
| Critical actions requiring approval | Creating a supplier record |
| Last test date and result | Date, pass rate, open issues |

## How Opencontroller does this

Opencontroller connects to SAP and other agent platforms (Salesforce, Databricks, HubSpot) and keeps a governed registry of agents, versions, tools and MCP servers across the estate, with delegated on-behalf-of identity, Posture for risky permission and configuration findings, multi-turn and red-team evals, and policy as code. Registration is not runtime takeover: control depth depends on the connector and traffic path. To see it with your SAP landscape, [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Do Joule Studio agents get their own backend permissions?**
SAP describes them as inheriting identity and principal propagation, so they act within the user's authorisation.

**Where do Joule Studio agents run?**
On SAP AI Core, with metering, tracing and security, and they become available in Joule once deployed.

**Do we still need a separate inventory?**
Yes, if you run agents on more than one platform. The Joule catalog covers Joule agents only.

## Key facts

Custom agents built with the agent builder in Joule Studio (part of SAP Build) run on SAP AI Core with built-in metering, tracing and security, and once deployed are available in Joule and registered in its catalog, according to SAP's architecture guidance. They reach SAP and non-SAP systems through APIs, SAP Build Process Automation workflows and MCP connections configured in SAP BTP, and are grounded on documents through SAP AI Core. They inherit SAP Cloud Identity Services and principal propagation, so actions stay within the user's authorisations. Governance focuses on who may build, which tools agents can reach, grounding data, release review, and an estate-wide registry. Opencontroller by Lyzr connects to SAP and keeps agents in one governed registry. Pricing starts from $250K per deployment per year.

## Related

- [How do you govern SAP Joule agents?](https://opencontroller.lyzr.com/blog/govern-sap-joule-agents.md)
- [How do you govern model access through SAP generative AI hub?](https://opencontroller.lyzr.com/blog/sap-generative-ai-hub-governance.md)
- [How do SAP AI Core resource groups isolate AI agent workloads?](https://opencontroller.lyzr.com/blog/sap-ai-core-resource-groups-for-agent-isolation.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
