Projects

What we have built — and run.

The overview for decision-makers and buyers: what each system is, which problem it solves, what it does — and the link to its own site for details, trials and terms. Every project here is live, in pilot operation or an openly documented research effort. Anything we cannot substantiate is not on this page.

Platform
Live

Verifact

Verifiable legal knowledge for AI assistants. Every answer with an official citation and a signed proof — or an honest gap.

Solves: Invented paragraphs and superseded statute versions in AI answers — a liability risk for law firms and legal tech.
- Official citation and confidence value with every answer
- Signed proof, verifiable offline by third parties
- Connects to Claude, ChatGPT and your own agents (MCP)
For: Law firms, tax advisories, legal-tech vendors and compliance systems that want to make AI assistants reliable

Project page verifact.network ↗
Platform
Live

Suveri

The sovereign AI workspace for companies in Europe. Open models in German data centres, every answer with model, cost and sources.

Solves: Using AI as easily as ChatGPT without handing company data to US providers.
- Chat, document analysis, research and verified law in one workspace
- Open models in German data centres, no storage of inputs
- Every answer shows model, cost and sources
For: Companies and teams in the EU that want to use AI productively under compliance requirements — especially in law, analysis and technical applications

Project page suveri.de ↗
Platform
Live

Kompetenzanker

AI literacy for the whole workforce — without jargon, with a credential that third parties can verify. EU-sovereign on STACKIT.

Solves: AI tools without understanding, shadow IT and no evidence of AI competence (EU AI Act Art. 4).
- Free foundation for the whole workforce, in-depth Champion Track
- Signed, publicly verifiable competence certificate
- Works-council-safe: aggregated metrics only
For: Mid-sized and large companies in the DACH region that want to build AI literacy internally and demonstrate it to auditors, customers or insurers

Project page kompetenzanker.de ↗
Platform
Live

OneLog

IT management with AI agents that resolve standard cases themselves — hosted in Germany, with a tamper-evident audit log for the evidence.

Solves: Five to eight separate IT tools and NIS2 evidence compiled by hand.
- One platform for inventory, updates, monitoring and tickets
- AI agents resolve standard cases on their own
- Post-quantum-signed audit trail for evidence
For: Managed service providers (MSPs), mid-sized companies subject to NIS2, KRITIS (German critical infrastructure) operators and public administration in the DACH region

Project page onelog.pro ↗
Partner integration
Live

Infercom Router

EU-sovereign AI models directly in Claude, ChatGPT and the command line — one connector, one login, cost per call visible.

Solves: Making EU-sovereign AI models available in the assistant you already use — without building your own application.
- One connector for Claude, ChatGPT and Gemini CLI
- Automatic model selection by task and cost limit
- Cost per call visible, no storage of content
For: Companies that already use Claude or ChatGPT and want to delegate sensitive tasks to EU models; partners who want to build sovereign inference into their own assistants

Project page suveri.de ↗
Platform
Live

TMT MentalTech

A coaching methodology as an app — to production in 17 days, without a single byte of client data being able to reach a US cloud.

Solves: Scaling a coaching method digitally without sensitive client data ending up in foreign clouds.
- Web and iOS app from one codebase
- Encrypted conversations, Art. 9 GDPR consent via touch signature
- Operated in Germany, no US components
For: Entrepreneurs and executives in the DACH region who want to work on themselves with a structured method

Project page zalana-mentaltraining.de ↗
Tool
Open source

HUGIN

An AI workspace that runs entirely on your own machine — and would rather say “I don't know” than invent something plausible.

Solves: AI assistants that send knowledge to foreign clouds and invent answers without a basis.
- Runs entirely on your own machine
- Honest abstention below the verification threshold
- A second model challenges every answer before you see it
For: Knowledge workers and teams who want to search their notes, files and documents with AI without anything leaving the machine

Project page
Tool
Internal pilot

SmartScan

Capture incoming post locally, extract it, attach deadlines and file it signed — on a single machine, without cloud. Our internal tool.

Solves: Incoming post with deadlines nobody may miss — without a cloud service for sensitive mail.
- Scanner capture and extraction via a local vision model
- Deadlines calculated in tested code, not by the model
- Signed log of every step
For: Ourselves, for now. The pattern fits any office that wants to process post digitally, locally and verifiably.

Project page

Research

Efforts without a product claim. We document openly what works and what does not — that is the foundation for everything that later goes into production.

Research
Research

EvoLab

A re-implementation of the architecture behind FunSearch and AlphaEvolve — a language model as mutation operator, a hard evaluator as selection. Local, without API keys.

Project page
Research
Research

AeroForge

A fully local design pipeline in which AI agents derive an aircraft configuration from physics and task alone — and which refuted its own assumptions.

Project page

A comparable challenge on your desk?

Send us three sentences. We reply within one business day with an honest assessment — including when we are not the right partner.

Book a 30-minute call Our method