No data center, no million-dollar budget — just consumer hardware. The result is a modular, fully local assistant: it knows what it knows, stays silent when unsure rather than guessing, researches missing knowledge itself, learns without retraining — and backs every answer with a source.
🔒 Runs entirely on your machine — no cloud, no token costs, your data stays in-house. I don't promise anything I can't show.
I'm retired — and instead of playing golf, I build AI. For about a year now I've been pursuing a simple idea on my own home PC setup: artificial intelligence should be honest, local and in your hands. No hype, no data leaks, no model a corporation can switch off from afar.
Language models guess brilliantly — but when they guess wrong, they sound just as confident as ever. For a poem that's fine; for an assistant that schedules appointments, checks invoices or answers questions about documents, it's a deal-breaker.
That's why my assistant has three properties I valued more than size: it is right when it's sure. It stays silent when it's unsure — “I don't know” is a valid answer, not a failure. And it researches and learns on its own when something is missing — permanently, with a source, without ever being retrained.
Why should you believe a pensioner working on something the big players are still tackling? You really shouldn't — so see for yourself: in an open comparison my AI takes on well-known models (you can replay the duels), and it even runs on a Raspberry Pi — answering a question in about half a second. What came out of it surprised me too.
Everything I build shares one idea: honest, local AI — that shows instead of claims.
One honest core system — and several things built on top of it or alongside it. Status marked openly.
Fully local, with hundreds of practical capabilities — from documents and email to research, images, audio and video. What's special isn't how much it can do, but how reliably: it doesn't guess, it cites. Modular by design — knowledge and capabilities are added as modules and are available immediately, without any retraining.
All prices are one-time — buy once, use forever. No subscription. (Shop & installer are in German for now.)
BRAIN for regulated environments: every statement with source and date, every step traceable, personal data provably erasable — matching the transparency requirements of the EU AI Act and the GDPR.
Assemble your own AI from modules — exactly as you need it. Knowledge and capabilities are added as modules and are available immediately, with no retraining. So you only pay for what you actually use.
Generates the technical EU AI Act documentation (Annex IV/XI) for an AI model — automatically, in a single run, fully locally at the customer's site. The model weights never leave the machine.
A fair head-to-head: BRAIN competes in twelve disciplines against large language models — from the 7-billion to the trillion class. Expected results are written down before each match and archived as a protocol. In the current duels BRAIN comes out ahead — with 0 hallucinations. Don't take my word for it: watch the replays.
A searchable knowledge base for retro computing: 124 platforms and programming languages — manuals, source code and tutorials — in a full-text search across millions of passages, each with its source.
A self-written Amiga operating system with a modern Workbench: window management like a present-day taskbar — plus a built-in ki.library that brings the AI right onto the classic Amiga.
Bundles BRAIN's office capabilities: email, appointments, invoices and documents — with a traceable rationale instead of a black box. Behind safety gates: reading automatic, sending only with approval.
Write to me, or stay up to date per project via the newsletter.
Prefer to write directly? ki@webkreators.de · Contact in writing only. Only your email address and your message are processed (see Privacy).
Choose which projects you'd like updates on — unsubscribe any time.
Only your email address + project selection, unsubscribe any time (see Privacy).
