SAP BTP Development and Integration

AI agents inside SAP S/4HANA

Most "AI for SAP" demos stop at a chat window that summarises a screen. We build agents that do the work — safely, inside your authorisation concept.

A Torotech agent is a small, auditable service on BTP that:

• reads business context through ABAP CDS views and OData V4 services (no direct table access, ever)
• reasons with a model of your choice (Anthropic Claude, Azure OpenAI, or SAP AI Core)
• summarises data, forms a query, compares records, or proposes an action — entirely through the app's own OData services, with the request shown to an approver before anything writes back
• writes back only after approval rules pass, and logs every step to a change document

Typical first projects

Process	What the agent does	Guardrail
Sales order holds	Reads block reasons, credit exposure and delivery dates; drafts the release or the customer note	Release above a threshold needs a human
Material verification	Compares inbound ASN, PO and goods-receipt data; flags mismatches with a suggested resolution	Never posts a GR on its own
Returns triage	Classifies return reasons from free text and photos; proposes disposition and credit	Credit memo is always approved

BTP extensions & integration

Every customisation you put in S/4HANA is a tax on the next upgrade. We build the things your business needs next to SAP on BTP — and connect them with events, not batch jobs.

• CAP services in TypeScript with proper authorisation, draft handling and OData V4 exposure
• Integration Suite flows for EDI 850/856/810, cXML punch-out, and 3PL shipment feeds
• Event Mesh subscriptions so BTP apps react to SAP business events in seconds
• Cloud Connector & Destinations configured once, documented, and reproducible with Terraform
• CI/CD with MTA builds, Cloud Transport Management and GitHub Actions

Reference architecture

S/4HANA ──(events)──▶ Event Mesh ──▶ CAP service ──▶ HANA Cloud
   ▲                                       │
   └────────────(OData via Cloud Connector)◀┘

Fiori & UI5 applications

We build Fiori apps with the same care we'd give a consumer product — clear flows, quick loads, keyboard and scanner friendly.

• Fiori Elements (List Report, Object Page, Analytical List Page) driven by CDS annotations and RAP behaviour definitions
• Freestyle UI5 / TypeScript for scanning, planning boards, blotters and anything a template can't express
• Launchpad & Build Work Zone design: catalogs, spaces and pages that match how teams work
• AI-assisted UI: inline explanation, smart defaults and natural-language filters powered by the same agents above

How an engagement runs

1. Map the workflow (same day). We sit with the people who do the work today and identify which OData services and entities the app already uses — and the exceptions that actually matter.
2. Build and wire it (1–2 days for a simple app). Chat shell, orchestrator, narrow LLM, OData calls — wired into the app that already exists. Nothing gets rebuilt underneath it.
3. Test it with the people who'll use it. Real phrasing, real edge cases, before anyone else sees it.
4. Go live. One integration, live for every user immediately — write-backs sit behind an approval inbox until the numbers earn wider auto-approval. More complex, multi-system workflows take longer, but the pattern itself doesn't change.

Stack

SAP BTP Cloud Foundry or Kyma · CAP (Node.js/TypeScript) · SAP Cloud Connector · Destination service · Integration Suite · Event Mesh · Anthropic Claude / SAP AI Core · Fiori Elements & freestyle UI5.

We've run this pattern in food distribution, healthcare and manufacturing, on both ECC and S/4HANA.