AI for SAP · Spyvee project

An ABAP agent that
proposes, never presumes

A development agent docked inside SAP ADT. It reads the ABAP you have open, answers in the context of your actual code, and hands back changes as a diff you approve or reject — so the AI does the drafting and you keep the last word.

Why this exists

ABAP moved on. Most ABAP codebases didn't.

Modern ABAP looks very little like the ABAP most systems are running. Inline declarations, table expressions, CDS, RAP — each arrived years ago, and each is still unfamiliar to teams whose day job is keeping decades-old code alive.

So the knowledge gets looked up somewhere else: a browser tab, a colleague, or a general-purpose AI tool that has never seen the code you're working in.

This agent lives in ADT and reads your open files, so its answers are grounded in your actual objects. Every change comes back as a diff you check before it lands.

Generic AI chatNever seen your system or your release
Guessing
Copy-paste into a browserOut of context, and back again by hand
Detour
Agent inside ADTYour open files in, a reviewable diff out
In place
What it helps with

The parts of ABAP people actually get stuck on

Modern ABAP syntax

Inline declarations, string templates, table expressions, CONV/VALUE/REDUCE — the modern constructs that older codebases and older habits often skip.

CDS views

Building and reviewing Core Data Services views — annotations, associations, and the modeling decisions that are hard to undo later.

RAP

The RESTful Application Programming model: behavior definitions, projections, and the wiring behind a working RAP service.

Debugging

Working through what a program is really doing when the runtime behavior and the code you're reading have stopped agreeing.

Performance

Spotting the SELECT in a loop, the missing index, the internal table read that turns linear work into quadratic.

Unit testing

ABAP Unit tests for code that was written before anyone was testing it, and advice on making untestable code testable.

How it works

Ask, review, apply

01

Ask, in ADT

Describe what you want in plain language, in a panel docked next to your editor. Tick “Include open files” and the agent reads the code you're actually looking at.

02

Review the diff

Proposed changes come back as a diff, not a wall of text to copy-paste. Read it inline, or open the full diff to see everything in context.

03

Apply — or don't

Apply the change and it lands in your editor. Reject it and nothing happens. Saving and activating stays your call, in ADT, as always.

Guardrails

What it can and can't do

These constraints aren't settings you have to find. They're how the tool behaves out of the box.

You approve every change

Edits arrive as a diff behind an Apply button, so nothing lands in your code until you say so.

It can't save or activate

Applying a diff updates the editor buffer. Saving and activating stay in your hands, in ADT, exactly as they are today.

You choose what it sees

“Include open files” is a checkbox, off until you tick it — your source is only sent when you turn it on.

No surprises about where it runs

The panel states plainly that messages go to an external AI provider, warns against production data and credentials, and names the provider and model in settings.

Where it fits

Part of Spyvee's SAP work, not a bolt-on

This is our AI for SAP practice pointed at the developer's side of the house — the same thinking behind our in-app assistant for SAP Fiori, applied to the people writing the code rather than the people using it.

  • Installs into the ADT you already use — no separate IDE, no change to how your transports and activation work.
  • Fits review-heavy shops: every change is a diff, so it slots into existing habits rather than fighting them.
  • Shares its foundations with the AI Agent for SAP Fiori and the rest of our AI-for-SAP work.

Want to see it against your own ABAP?

Tell us what your team is working in — the release, the codebase, the parts that slow you down.