On-prem AI for SAP/ABAP · built on Claude Code

Built by an ABAP developer who didn't trust AI either.

For years the experiments disappointed — every model needed re-tuning, and it was usually faster to just write it yourself. In 2026 that changed. So I built the tool I'd actually use: a self-hosted agent that knows your standards, works inside your SAP via ADT, and takes the parts that used to eat your hours.

Honest start

If you've tried AI on ABAP and it let you down — same.

I spent years testing models on real ABAP work. Most of the time the honest call was: cheaper to write it myself than to check and redo what the AI produced.

What changed in 2026 isn't hype. Capable models plus agents crossed the line where handing off real work finally pays off. But a raw model still needs setup, rules and context before it's useful on ABAP — and that's exactly the part nobody wants to do over and over. Makion is that setup, done: configured for ABAP, working to your standards, out of the box.

Not a chat box — a configured ABAP workflow.

Everything an ABAP team should do but rarely has time for — built in and enforced, not left to willpower.

Your standards, enforced

Best-practice rules and a shared methodology the agent always follows — your conventions, applied every time, across the team.

Roles for real tasks

Task roles for how work is actually done — including dedicated code-review roles — instead of one generic prompt.

A growing knowledge base

Your patterns, examples and templates collected once and reused across the project — the agent builds on what you already have.

Specifications, unified

Bring specs into one internal format, versioned and kept in the repo — a single source of truth, not scattered documents.

Tests that get written

Automate the unit tests almost nobody finds the time for — and the routine checks that have to happen on every change.

Reuse, not rewrite

Read before write: the agent reuses existing objects and proven patterns from your project instead of inventing new ones.

Inside Makion

A developer's workspace for SAP — not a chat window.

Makion — Chat
Makion — chat with task roles
Pick a role, describe the taskRoles for how the work is really done — development, architecture, code review, unit tests, ATC, research, specification — not one generic prompt.
Makion — Methodology
Makion — methodology and rules
Your method, always appliedBase rules plus a methodology per role that the agent always follows — edit it to match your standards.
Makion — Settings
Makion — per-project SAP connection
Per-project SAP connection, on-premEach project has its own gateway; credentials are encrypted locally and the connector links only to your development system.
For teams & companies

ABAP talent is scarce and expensive. Makion is ready today.

A strong ABAP developer is hard to find and expensive — and training your own is a long road. Makion gives the team you already have more throughput, with consistent quality, best-practice and code review built in — running on your own infrastructure, with one shared approach instead of everyone's personal style.

Runs on your server. Powered by your Claude Code.

Makion is your deployment — not a SaaS we host. You stay in control of where it runs and what it touches.

Self-hosted on your serverMakion runs on your own infrastructure. It's your deployment, not a third-party service holding your code.
Your own Claude Code subscriptionThe agent is built on Claude Code and uses your own subscription — your terms with Anthropic, no SaaS middleman in between.
Development systems onlyThe connector links to your SAP development system — never production. It's a tool for building, not for touching live data.
Credentials stay encryptedSAP credentials are encrypted locally (Windows DPAPI) and the connector runs on-prem; nothing about your SAP login is kept in plain text.

Install

Get the desktop app on the developer's machine.

Connect dev SAP

Start the on-prem connector to your development system.

Pick a project

Each project carries its own rules, roles and knowledge base.

Work with the agent

Describe the task; it works with real ADT tools to your standards.