Step 1: SET UP
One afternoon. Greenfield or brownfield, one abapGit export, your guidelines as checks you control, an owner for every program.
WHY NOW
ECC SUPPORTends in 2027.THE CLOCK RUNS.
SAP ECC 6.0 · MAINSTREAM
MAINTENANCE ENDS 2027 SAP
The usual move takes 9 to 18 months, paid in consultant days. Six in ten projects run over budget and schedule. Horváth, 2025
THE OLD WAY · THE NEW WAY
SAME GOAL.
NEW SPEED.
| Topic | Usual project | Greenswarm |
|---|---|---|
| Time | 9–18 months | About a week |
| Price | Consultant days, open-ended | One fixed price |
| Overrun risk | Yours | Ours |
| Evidence | Status slides | A trace from every finding to its sign-off |
TEN VERSIONS. ONE WINNER.
EVERY VERSION TESTED ON YOUR DATA AND IN REAL SAP
the best parts win.
HOW IT WORKS
FIVE STEPS.one click.
After the setup, the swarm works on its own. Your people review the result, not every step.
Step 2: TEN TEAMS BUILD
Ten AI teams, ten approaches, all outside your SAP system. Reviewer agents check every step.
- T1Change as little as possible
- T2Official interfaces only
- T3All database reads in one place
- T4Speed first
- T5Tests first, then the fix
- T6Clean, small classes
- T7Follow the new S/4 tables
- T8Guard every edge case
- T9Pure logic, thin database layer
- T10Follow SAP’s notes exactly
Step 3: TESTED, CHECKED IN SAP
Every version runs on your data, then on a freshly reset SAP system. SAP catches what fast tests miss.
Step 4: BEST PARTS COMBINED
The best version of each program goes into one build. That build is tested again, in SAP too.
- Stock movementsfrom Team 10
- Order pricingfrom Team 1
- Open deliveriesfrom Team 10
- Overdue invoicesfrom Team 1
- Material uploadfrom Team 1
- Stock dashboardfrom Team 10
Combined build: 26 of 26 business checks in SAP
Step 5: DOCUMENTED, SIGNED OFF
Code ready for normal transports. Docs for every program. A trace from each finding to its fix, test and owner sign-off.
The trace caught a gap every test had passed. The owner accepts it in writing.
WATCH IT RUN.in three minutes.
Recorded on our sample package. The swarm run is sped up.
WHAT YOU CAN HOLD US TO
FIVE
PROMISES.in plain words.
FAST.
About a week instead of 9 to 18 months.
CHEAPER.
One fixed price, below an integrator’s. Overruns are our cost.
66% lower AI running costs when we host open models ourselves, as long as the GPUs stay busy (Cline, Oct 2026).
AUDITABLE.
Every change traced to its finding, its test and its owner’s sign-off.
DOCUMENTED.
Functional and technical docs for every program, before and after.
PRIVATE.
Open models on your servers or in a German BSI C5 cloud. Our AI never touches your SAP.
COMPLIANCE
YOUR SAP
STAYS CLOSED.
- YOUR SYSTEMSAP ECCYour developer exports the code with abapGit.
- WHERE THE AI WORKSGREENSWARMOn your servers or in a German BSI C5 cloud. Open models. Every step recorded.
- YOUR SYSTEMSAP S/4HANAYour developer imports it, then your usual transports.
No AI inside your SAP: no add-on, no endpoint, no user for our agents.
PROOF SO FAR · 1–2 NOVEMBER 2026
WHAT WE SHOWED.honestly.
No customer results yet. Everything below ran on a sample package we wrote ourselves: six classic ECC programs, each hitting a typical S/4HANA break. Unchanged on S/4HANA, it passes 2 of 26 business checks.
- 190AI agents on one GPU
- Self-hosted NVIDIA Nemotron 3.5 Super, one B300
- 10 of 10teams passed our tests
- All 26 business checks
- 7 of 10versions passed in SAP
- On a freshly reset SAP system
- 26 of 26for the combined build
- Checked again in SAP
- $0.04per agent-hour
- When we host the model
- 26 of 26on a real ECC 6.0 system
- Our test system; export identical
SAP checks ran on SAP ABAP Platform 2025 (developer edition), reset before every version; a few S/4HANA tables there are stand-ins. Costs are measured from our runs and leave out our people’s time.
WHO IS BEHIND IT
15 YEARS
IN SAP.Mario Andreschak
Trained ABAP developer. SAP integration and finance transformation in EY’s IT Advisory practice, then independent. Built the first MCP server for SAP ABAP in January 2025 (credited as the first). Creator of FLUJO and FACTORY, the agent tools Greenswarm runs on.
QUESTIONS
ASKED FIRST.
No. Your developer exports the custom code once with abapGit, a standard open-source tool, and imports the finished code once. Everything in between happens outside your SAP.
The abapGit export, sample data for the business tests, your guidelines if you have them, and a business owner for each program who answers open questions and signs off.
Where you decide: your own servers, a German cloud certified under BSI C5, or, if you prefer, an AI provider’s API. With the first two, no outside AI provider sees anything.
The agreed scope, migrated, tested, checked in SAP and documented, up to your owners’ sign-off. If it takes us longer, that is our cost. The pilot is paid only on acceptance.
You see it, in plain words: the program, the reason, what it means for the business. The owner decides: we fix it, or the owner accepts it in writing with a plan.
Early, and open about it. The whole method runs end to end on a sample package we wrote, including checks on a real SAP system and a real ECC 6.0 system. A pilot shows it on your code.
THE PILOT
ONE PACKAGE.
ONE PRICE.pay when you accept it.
EMAIL US ABOUT A PILOT- Fixed price, agreed before we start
- About a week from your export
- Your code stays where you choose