Co-founder & CTO, VibeMind · Munich, Germany

VibeMind turns a founder's intent into coordinated, auditable agent work — across memory, tools and infrastructure — without requiring a technical background.

A solo founder does the work of ten people, with nobody to delegate to. Today's AI answers questions; it does not do the work. And the tools that come closest are built for people who already think like developers — which excludes exactly the people who need the leverage most.

VibeMind is a voice-first agentic operating system. You say what you need. Specialized agents carry it out against your real systems, share memory across sessions, and leave an auditable record of what they did.

I'd rather be checkable than impressive. Status as of August 2026:

What exists is a technically substantial integration prototype. A clean, reproducible release and proof of external usage are still ahead of me, and I don't claim them.

• ~4,000 early-access requests over nine months, with zero paid marketing — organic reach through the builder community and LinkedIn.
• 12 volunteer contributors have offered their time before there is any money, and are waiting to be onboarded through the product rather than through meetings. That is the first real test of the core claim on a group larger than its authors.
• No paying customers yet. The list demonstrates pull, not willingness to pay; establishing that is the next job, not a claim I make now.

Three co-founders, plus a volunteer community — deliberately kept as separate categories.

Plus twelve unpaid contributors — not employees, not founders, not counted as team.

Numbers below come from the space/agent registry in the core repository, not from a pitch deck. They describe code and configuration — not a passing end-to-end suite, and not production readiness.

Architecture documentation is public in Vibemind_V1 — component boundaries, request lifecycle, deployment planes. The product core itself is not open yet.

Verifiable execution. Agent actions are written to an append-only, cryptographically chained audit log with deterministic hashing, so any record can be independently checked and tampering is detectable. That property is the reason delegating work to agents on machines you don't personally supervise is a reasonable thing to do.

Where it runs is part of the product. VibeMind is built to run on hardware the user controls — a household node with ~24 GB of GPU memory serves open-weight reasoning models in the 27–35B class at 4-bit quantisation. Work that exceeds it routes to a larger open-weight model on GDPR-compliant EU infrastructure, never a US hyperscaler, and the routing decision is visible to the user. Europe should not have to choose between using AI and keeping its data.

Next. A trading space and a learning platform are in progress.

The Brain space (Tahlamus) is original PyTorch work rather than model wrappers — a bet that the coordination layer, not the model, is where systems like this break.

Not a model moat — operational depth. Making dozens of agents coordinate reliably against private data, with state that survives failure, is precisely where these systems break, and that is what the last eighteen months went into. Each user additionally accumulates personal memory and retrieval context, which makes the assistant more useful over time and more costly to replace.

AI and automation engineering; previously mechatronics at BMW, then applied informatics (bioprocess informatics) at HSWT Weihenstephan-Triesdorf. Working in generative AI since 2021. German and English.

Python · TypeScript · PyTorch · AutoGen · MCP · FastAPI · Electron · Three.js · Next.js · Docker · Proxmox · Supabase · Qdrant · n8n

• Website — vibemind.space
• Company — VibeMind on LinkedIn
• Email — felix@vibemind.space

Building agent infrastructure that runs on hardware you control.