01AI Automation

CustomAIagentsbuiltintoyourSAPlandscape

We build autonomous AI agents on SAP BTP and AI Core, integrate existing ones into your environment. Whether you want to automate a process or connect an agent you already have, we handle the build, the wiring, and the handover.

SAP AI Agent Development
/THE CHALLENGE

Most SAP customers know AI can unlock value in their processes, but face the same barriers: unclear entry points, unknown licensing requirements, and fear of long implementation timelines with uncertain ROI. Joule is powerful but requires the right landscape setup. Custom agents require deep expertise in both SAP architecture and AI frameworks. The gap between a compelling demo and a production-ready agent is where most projects stall.

/OUR APPROACH

We match the delivery path to what your landscape supports today. Three options cover the full spectrum: a low-code Joule Studio flow for organisations with the right BTP entitlements, a pro-code BTP multi-agent architecture for complex cross-system orchestration, and a Custom Fiori AI Chat for restricted or on-prem environments where Joule is not yet licensed. All three produce agents your team owns and can extend without us.

LOW CODE

Joule Studio - Visual Agent Builder

Build SAP-opinionated agents directly in Joule Studio without writing code. You describe your intent; Joule analyses your landscape, generates a product requirements document, produces a technical specification, and scaffolds working code - deployed to SAP-managed runtime in a unified flow. We configure, extend, and tune the output to fit your exact process and data model.

SAP Joule StudioSAP AI CoreYAML SkillsFiori Launchpad
Agent overview
Expertise & Instructions
Model Settings
MCP Servers
Tools & Output
Choose what happens next
Agent overview
Expertise & Instructions
Model Settings
MCP Servers
Tools & Output
Choose what happens next
Expand
PRO CODE

BTP Multi Agent - Custom Architecture

For complex, cross-system orchestration where custom reasoning logic is required. We build multi-agent systems using LangGraph and the A2A protocol on BTP Cloud Foundry, Kyma or Docker - routing agents, RAG pipelines, and tool-calling agents with direct access to S/4HANA OData, MCP servers, or OData APIs. Every component follows SAP Clean Core principles and lives side-by-side, never inside the core system.

BTP Cloud FoundryLangGraphA2A ProtocolMCP HubABAP AI SDK

PRO-CODE · BTP MULTI-AGENT

Python agents on SAP BTP — production-grade from day one.

Agent
LangGraphLangChain-LiteLLMA2A SDK
LLM / AI
SAP AI CoreGPT-4.1LiteLLM router
SAP BTP
HANA MemoryXSUAA / OAuth2Cloud Foundry
Ops
OpenTelemetryOTLP / ProtoDockerStarlette
A2A Protocol
MCP Hub
ABAP AI SDK
S/4HANA OData

ORCHESTRATION

LangGraph StateGraph — deterministic, testable, inspectable.

agent.py — _build_graph()

1def _build_graph(self):

2 tool_node = ToolNode(TOOLS)

3

4 async def call_model(state):

5 response = await self.llm.ainvoke(...)

6

7 builder = StateGraph(MessagesState)

8 builder.add_node("model", call_model)

9 builder.add_node("tools", tool_node)

10 builder.add_conditional_edges(...)

11 return builder.compile()

START

MODEL

call_model()

TOOLS

has_tool_calls

END

TOOLS & INTEGRATIONS

Six tools. Every action confirmed before it executes.

run_analysis_tool

SAP AIF OData · 42 error records

run_doc_error_catalog

SharePoint · AI Core vector search

draft_email_tool

Drafts report from Markdown

send_email_tool

Email Service · HITL confirmed only

HITL

draft_jira_tool

Drafts Jira Story from report

create_jira_tool

Jira REST API v2 · HITL confirmed only

HITL
1
Agent drafts
2
Human reviews
3
User replies "yes"
4
Action fires
Codemine Telemetry — OTLP trace · aif-analysis-agent
SPANDURATION
invoke_agent_span: aif-analysis-agent
450ms
├──LiteLLM: sap/gpt-4.1 (call_model)

gen_ai.request.model=sap/gpt-4.1

180ms
├──run_analysis_tool

aif.period=last 24h · records=42

95ms
├──LiteLLM: sap/gpt-4.1 (call_model)

tool_calls=run_doc_error_catalog

210ms
├──run_doc_error_catalog_tool

grounding.query=FI/001 Amount mismatch

88ms
├──LiteLLM: sap/gpt-4.1 (final)

gen_ai.usage.total_tokens=2847

175ms

450ms

total

2 847

tokens

6

spans

Tech stack
LangGraph
Tools & HITL
Observability
Expand
RESTRICTED

Custom Fiori AI Chat - On-Prem Path

For landscapes where Joule licensing is not yet in place or cloud connectivity is limited. We build a Custom Fiori AI Chat with an A2A Agent Cockpit that operates against on-prem HANA data - with a guaranteed migration path to Joule when the license arrives. Same agent logic, same tool integrations, different runtime.

SAP FioriSAP HANAA2A Agent CockpitREST APIs
/DELIVERY PHASES
01

Assessment

We map your SAP entitlements, landscape configuration, and process candidates. The output is a clear entry point - Joule Studio, BTP multi-agent, or Fiori AI Chat - matched to what you actually have licensed today.

02

Design

Agent architecture defined before any code is written. We specify tool integrations, data connections, reasoning flows, and success criteria - so you approve the blueprint, not the result.

03

Build

Structured development following SAP best practices. API connections wired, business logic assembled, and the complete solution package brought together - with observability and error handling built in from the start.

04

Validate

Automated unit tests and AI-powered scoring verify the agent behaves against your original requirements. Gaps are caught and closed here - not after go-live.

05

Deploy

Agent deployed to your SAP-managed runtime. End-to-end validation, and a structured hypercare window. Your first users interact with a production-ready solution, not a prototype.

06

Improve

Monthly performance reviews, usage analytics, and iterative enhancements. We transfer full ownership to your team - and stay available to keep the agent sharp as your business evolves.

/MORE SERVICES
All services →