AICore Hub
📋 Requirements & Implementation

What it takes to run it.

This is the practical companion to the overview. It sets out exactly what you provide, what we deliver, how it installs into your existing SAP system, and how it stays under your control — no story, just what you need to plan a rollout.

💡
The overview page tells the story and shows the product in motion. This page is the checklist behind it — aimed at whoever has to scope, approve, and stand up a pilot.

What you provide

AI Core Hub supplies the control room, the assistant, the knowledge engine and the skills. To switch it on, you bring the following.

1

Models — the brain and the librarian.

Two model roles are needed, and they are different types. A generative model (the brain) reads context and the screen, reasons, calls skills, and writes the answer — essential. A small embedding model turns documents and questions into vectors so the knowledge base retrieves by meaning — this powers grounded answers. Both can be a private model you host, a hosted endpoint, or a corporate AI gateway; usually one provider serves both, so this is one relationship, not two. Optional: split the brain into a fast text model for everyday questions and a stronger reasoning model for the hard cases, with the Hub routing between them.

2

Zones — a team per product.

A small, dedicated team responsible for one product or area. They own its documentation, decide its agents and skills, and keep its knowledge current. Separating products into zones keeps answers accurate and keeps the cost per question (the AI tokens) low — an agent only ever reads its own zone. You stand these up around the teams you already have.

3

An existing SAP system.

A system to load it into (ECC or S/4HANA). No SAP BTP, no separate platform, no extra servers — it runs inside the system you already operate.

4

Your documents (per zone).

The manuals, specs, process notes and resolved tickets each team wants its agents to answer from. Optional at the start — the baseline-documentation service (below) can give every agent a head start before a team writes anything.

What we deliver

🎛

The Hub (control room)

Where teams set up zones, applications, agents, skills, keys and the knowledge base — loaded into your existing system.

💬

The in-screen assistant

The docked chat agent, reachable from any screen, that reads context and answers from your knowledge under the user's own authorisations.

Skills & the engine

The ready skills (screen reading, safe queries, field help and more), the grounding/knowledge engine, and the framework to add your own.

🛠

Setup & a guided pilot

Installation into your system, the working configuration, and hands-on guidance through a first pilot zone.

How it installs

AI Core Hub is delivered as a set of standard objects that load into your established SAP functionality. There is no separate runtime, middleware or cloud platform to provision — the configuration Hub and the in-screen assistant simply become available within the system you already operate. The only outbound connection is from your system to the model you nominate, over one managed channel with keys held centrally. Every data read happens under the user's own SAP authorisations.

A typical adoption path:

  1. Connect your models. Register the generative + embedding endpoints and keys centrally.
  2. Baseline documentation (optional). We document your system so agents start with knowledge from day one (see below).
  3. Stand up a first zone. Pick one team and one application as the pilot; load its key documents.
  4. Configure an agent. Define a persona, choose its skills, dock it into the chosen transaction.
  5. Prove the value. Let real users ask real questions; measure time saved and ticket deflection.
  6. Expand. Add zones for other teams, then introduce workflow agents where a process can run on its own.

Security, control & governance

  • Your data stays in your system. Answers are built from your documents and tables; only the prompt you permit travels to the model you chose.
  • Authorisation-respecting. Every data read runs under the user's own SAP authorisations — the assistant never surfaces data the user could not already see.
  • Read-only by default. Free-form data access is read-only; changing data happens only through skills a team has deliberately built and validated.
  • Scoped capability. Skills are governed at global, zone and application level, so each agent can do exactly what it should — and nothing more.
  • Central key management. Model keys are resolved centrally, never scattered across users or stored in the open.

Optional: baseline documentation of your system

Most SAP landscapes have thousands of objects and very little documentation. As an optional part of implementation, we scan the objects across your system — tables, programs, transactions, custom code, and the way they connect — and produce a system-wide picture of how it fits together. This becomes baseline knowledge every agent can draw on, so even a zone that hasn't loaded its own documents still answers usefully from day one.

It is delivered as a single comprehensive document, or, if you prefer, split into separate documents per area or product. The result is yours to keep — living documentation of a landscape that usually has none — and can be refreshed as your system changes.

No cold start

Agents are useful on day one, before any team has documented a thing.

🔗

Relationships, not just lists

It captures how objects connect, so answers understand context rather than just names.

📈

A living baseline

Team documents layer on top and take priority; the baseline stays as the fallback.

Roadmap

CapabilityStatus
In-screen chat agents (personas) with knowledge + skillsin the prototype
Per-team zones, applications, skills registry, key resolutionin the prototype
Knowledge base (grounded answers from your documents)in the prototype
Baseline documentation of your whole system (one or many documents)service
Workflow agents (ServiceNow responder, intelligent background jobs)rolling out
Guided data changes inside standard SAP transactionsroadmap

Key terms

TermMeaning
ZoneA dedicated team that owns one product or area, with its own agents, documents and keys.
ApplicationA single transaction or a whole business area — wherever a group of people do their work.
Agent (persona)A name, an instruction and one focused job. Never a generic “bot”.
SkillA concrete, safe action an agent may take inside SAP — read the screen, check stock, run a read-only query.
Knowledge baseA zone's documents, filed by meaning, that agents answer from (RAG).
KeysThe connection to your AI models, resolved centrally at company / team / user level.
Chat agentA persona that answers questions, docked beside a screen.
Workflow agentA persona triggered by an event or schedule that runs a process on its own.

Next steps

The fastest way to evaluate AI Core Hub is a short pilot on one team and one transaction.

  1. Watch the demonstration on the overview page.
  2. Identify a pilot team (zone) and a transaction where users get stuck most often.
  3. Confirm the models you want to connect.
  4. We set up the zone, load its documents, configure an agent, and dock it into the chosen screen.
  5. Measure the result with real users over a couple of weeks.
AI Core Hub · an AI core that lives inside SAP · part of the rush-team.dev family ← Back to overview