Skip to main content

A factory for Joule agents.

A catalog you can read today. A factory you can use next week.

Production Joule agents on customer S/4HANA, built in our factory and pointed at your live system through a governed agent gateway so they reason against real business data, not a copy of it. The directory keeps growing as new work surfaces.

Talk to a builder See the catalog
SAP JOULE AGENTS FIN SCM INT EWM SRV MCP SERVERS · PER FUNCTIONAL AREA SCOPED TOOLS, EXPOSED OVER ODATA MCP MCP MCP MCP MCP SAP BUSINESS AI PLATFORM BTP BDC AI FOUNDATION SAP S/4HANA CUSTOMER SYSTEM OF RECORD FINANCE · SALES · SCM · WAREHOUSE · SERVICE MINDSET ASSEMBLES THE STACK
The Mindset team celebrating their SAP BTP Hackathon win
SAP BTP Hackathon: won by the same team that builds the agent factory. SAP BTP Hackathon
Mindset team members on stage at SAP Sapphire presenting the Twenty Joule Agents launch
On stage at Sapphire: twenty agents, validated by SAP. SAP Sapphire 2026
Why this is different

The interface to SAP is changing.

For fifteen years the interface to SAP was a screen, then it was a Fiori tile. Now, increasingly, it's a conversation. Joule is SAP's AI agent layer. A Joule agent reads your business in S/4HANA, acts on it through governed tools, and explains itself in plain English.

That sounds simple. It is not. The agent only works when the platform underneath it is clean, the data layer underneath that knows what your business actually means, and the integration around it is wired for tools to be called safely. We've been building that stack since 2010. The Factory is what it turned into.

The agent might live in Joule. It might live in your AWS Bedrock, OpenAI, or Anthropic stack. The SAP-side work is the same. The MCP layer still has to expose the right tools over OData. Business Data Cloud still has to ground them. The Integration Suite still has to broker them safely. That's the work we do. We are your SAP AI partner either way.

And before any of that, you have to aim the agent at the right work. We mine the real process in SAP Signavio and map the systems it runs on in SAP LeanIX, so the agents we build land on the steps that actually cost you, not the ones that demo well. That is our process intelligence practice.

How we build

Five steps from blank slate to running agent.

Same pattern, every time. The factory works because the steps are honest, not because the demo is shiny.

  1. 01

    Design workshop

    Half a day in our SAP AppHaus. Frame the use case, the user, and the truth the agent needs.

  2. 02

    Joule Studio build

    Joule Studio + Claude + SAP Domain Models. Agent scaffold up in days, not quarters.

  3. 03

    Trust layer

    Ground the agent in BDC and governed OData, with tools scoped per MCP server. So it works from what the data actually means, not guesses.

  4. 04

    A2A and MCP wiring

    Tools scoped through MCP servers. Agent-to-agent calls via the A2A protocol.

  5. 05

    Production on S/4HANA

    Cloud Connector tunnels to your live system. Identity governed by Joule. No data copy.

In the catalog

Agents across every SAP function.

Every name below is a deployable solution in our agent directory, about two dozen today, twenty of them validated by SAP at Sapphire 2026. It keeps growing as customers and our builders surface real work the platform already knows how to do.

Featured agent Ready to deploy

Delivery Block Resolution.

Sales orders get blocked every day. Finance has a view, credit has a view, inventory has a view, the customer has a view. The agent gathers them, reasons across the four domains the way an experienced order specialist would, and recommends the unblock with the supporting evidence, which saves the back-and-forth that used to eat a day.

Talk to the builder

  • Finance

    • Invoice Discrepancy Reconciler
    • Credit Management
    • GL Period-End Close
    • Intercompany Reconciliation

  • Sales & CX

    • Delivery Block Resolution
    • Sales Order Exception
    • Customer Returns

  • Procurement & Supply Chain

    • PR-to-PO Accelerator
    • Procurement Risk
    • Order Supplier
    • MRP Exception

  • Warehouse

    • Warehouse Order Agent
    • Warehouse Receiving Agent
    • Warehouse Shipping Agent

  • Service & Maintenance

    • Plant Maintenance Service Technician
    • Plant Maintenance Work Orders

  • Data & Analytics

    • BDC Admin Agent
    • BDC Story Creation Assistant

  • Fiori & Platform

    • Fiori App Analyzer
    • Fiori Support Agent
    • Integration Suite Helper

The directory keeps growing across every function our customers run on SAP. This is the home for the catalog. New agents publish here as they ship.

What sets the Factory apart

Most agents are demos. Production agents are a different sport.

What demos do

  • Copy data into a vector store and reason against a snapshot.
  • Show one happy path against a sample dataset.
  • Stop at the workshop or the conference booth.
  • Treat identity, governance, and tool scope as details.

What production agents do

  • Stay in the source

    Read and write through OData, with human approval on the writes that change your system. The customer's S/4HANA stays the truth.

  • Survive real workflows

    Act on live transactions with real users, real edge cases, real exceptions.

  • Ship and keep shipping

    New agents land in the catalog as the work surfaces them. The ones already there evolve as the work does.

  • Take identity seriously

    Every action runs as a real user with real S/4HANA roles. Tools are MCP-scoped, not god-mode.

Critical insight

If the agent doesn't have to live on a real system, you're watching a demo.

Trust by design

Your data stays put.

Trust is the boring part nobody demos. It's also the part that decides whether the agent ever leaves the lab. Four things hold ours up.

No copy

OData tunnels to the live system.

Read and write through OData V2 and V4. No nightly extract. No shadow data store to keep in sync.

Encrypted transport

SAP Cloud Connector at the edge.

Every call is tunnelled through the customer's own Cloud Connector. The agent reaches in, never the other way around.

Identity governed

Joule owns the user context.

Every action runs as a real user with real S/4HANA roles. No service account that can do anything.

Scoped tools

MCP servers, narrow on purpose.

Each domain ships its own MCP server. An agent can only call the tools it has been handed.

Proof in the work

Agents only work on a clean platform.

We've built that platform for some of the largest SAP customers in the country. Joule agents land on top of work we've been doing for years.

Automotive

Lucid Motors

65 interfaces off deprecated CPI NEO, zero downtime

Read the work Energy and Utilities

Border States Industries, Inc.

Border States: Gen AI Sales Assistant on SAP BTP

Read the work Consumer and Retail

The Clorox Company

How Clorox built the foundation for autonomous finance on S/4HANA and Datasphere

Read the work Food and Agriculture

CHS Inc.

How CHS unified its SAP estate on BTP, from a single launchpad to identity governance and AI

Read the work
All case studies
Lessons from production

Six things we know now that we didn't six months ago.

The Factory taught us a few things the demos never could, the kind of lessons that only show up after the agent is running on a real customer system at 3am.

  1. On speed 01

    We use agents to build agents.

    ABAP through SAP-ABAP-2. Test generation by agents. Code review with Claude in our delivery flow. The catalog grows weekly because the Factory itself runs on AI.

  2. On the work 02

    The data layer takes longer than the agent.

    Most of the lift is BDC, governed OData, and the MCP tool layer. The LLM is the easy part. Customers who skip the data layer ship demos, not agents.

  3. On security 03

    Identity is the unsung work.

    Joule's role-based execution matters more than the model choice. Get the user context wrong and the agent is unsafe. We spend more time here than people expect.

  4. On what users want 04

    Trust beats benchmarks.

    Production users want recoverability and explainability. Leaderboard scores do not ship. The agents that stay in production are the ones the user can correct.

  5. On use cases 05

    Not every workflow wants an agent.

    The ones that do want very different agents. Picking the right use case is half the work. We have walked away from agent ideas, and the catalog is sharper for it.

  6. On the platform 06

    The Factory only works because the platform did.

    Clean BTP and clean S/4HANA are the precondition for agents that survive. We have been doing that platform work since 2010. That is why the catalog ships.

From the show floor · May 2026

We shipped the Factory at SAP Sapphire 2026 in Orlando.

Real Joule agents on real S/4HANA, in the booth, with the customers using them in the room. The microsite that launched at Orlando is still up as the archive of what shipped that week. The catalog has kept growing since.

Sapphire 2026 archive
From the build

Real rooms. Real customers. Real agents.

The Factory does not run on slideware. These are the sessions where the agents actually get built, with the people who use them, on the systems they run on.

A design session in our SAP AppHaus, Minneapolis

AppHaus · workshop

A design session in our SAP AppHaus, Minneapolis

On stage with a customer at their executive summit

Customer · summit

On stage with a customer at their executive summit

Sharing the work in progress with the room

Workshop · review

Sharing the work in progress with the room

How to engage

Three paths in. Builders on every one.

From a first session to a sustained SAP AI partnership. Whatever stage you are at, you meet our builders, not a pitch team.

Path one ~ half a day

Workshop with a working agent at the end of it.

We come to your office, sit with the SMEs, frame a use case the catalog can fit, and stand up a working agent against a connected dev or test system before we leave, with the path to production mapped. The fastest way to know whether a Joule agent belongs in your stack.

  • Use-case framing in the room, not afterward
  • Agent runs against a connected dev or test system, with the path to production mapped
  • You leave with the agent and a next-week plan
Path two Production build

Build the agent your team will live with.

When the workshop turns into something you want in production, our builders take it the rest of the way. Trust layer wiring, identity governance, A2A and MCP integration, deployment on your BTP, and a runbook for the team that owns it.

  • Direct line to the engineers shipping the catalog
  • Trust layer grounded in BDC + governed OData
  • Handover that doesn't leave a black box behind
Path three Strategic partner

We become your SAP AI partner.

For customers running an AI roadmap across their SAP estate. Named senior partner from our team. Quarterly architecture reviews. Cross-portfolio learning so what shipped for another customer accelerates yours. Aligned with your AWS, OpenAI, or other AI choices, not against them.

  • Dedicated senior partner as your AI counterpart
  • Quarterly platform + roadmap reviews
  • First look at the catalog as it grows
  • Lessons learned shared across your team
Related: MIND, our productized agents SAP AppHaus, Minneapolis Business Data Cloud SAP Business AI
Recent thinking

Written from the build.

All insights
Joule & Agents JUN 11, 2026
SAP SIT Retrospective

SAP SIT Retrospective

I recently got an opportunity to attend SAP SIT for the first time at the SAP Campus in Whitefield, and it was a truly rewarding experience....

Read the post
Funded by SAP · through 2026

SAP is funding this work right now.

SAP put €100M (~$114M USD) behind partner-led Business AI adoption, up to €100,000 per customer. We're an approved partner and we guide you through SAP's process. The window closes at the end of 2026.

See if you qualify

Bring us your toughest agent.

The first call is a working session, not a pitch. Tell us where SAP slows your people down and we will sketch the agent with you.

Talk to a builder