Ask MindMap — a sovereign RAG demo you can talk to.

Ask about sovereign AI, the EU AI Act, agentic AI or RAG. Every answer is drawn only from MindMap's own corpus and cites its source — the exact grounded pattern we deploy on your documents, inside your perimeter, with nothing leaving the box.

Ask MindMap

Grounded RAG demo · answers from our own corpus · no data egress

Ask me anything about sovereign AI, the EU AI Act, agentic AI, RAG, document intelligence, or how MindMap engages. I answer only from MindMap's published knowledge base, and I cite every source — the same grounded pattern we deploy on your documents, inside your perimeter.

How this works

Retrieval in, grounded answer out — nothing leaves

Retrieval over a private corpus

Your question is matched against a fixed knowledge base — here, MindMap's own FAQ and glossary. In a deployment, it is your documents: policies, contracts, product manuals, the knowledge that cannot leave the building.

Grounded answer, every source cited

The answer comes from the retrieved passages, not from a model's memory, and every claim links back to its source. If the corpus does not contain it, the system says so rather than guessing.

Nothing leaves the box

This demo makes no call to any external LLM — retrieval runs entirely on our own server. Your deployment runs the same way inside your perimeter: air-gap capable, logs in your SIEM, zero egress.

How sovereign RAG works → Meet ChatNext →

FAQ

About the demo

Is this using ChatGPT or Claude under the hood?

No — and that is the point. The demo answers by retrieving from MindMap's own published corpus with no call to any external model, which is exactly how a sovereign deployment behaves: the model and the data stay inside the perimeter. It illustrates the retrieval-and-grounding pattern; a production deployment adds a sovereign open-weights LLM (Llama, Mistral, Qwen, or one of the MindMap model families) to phrase the answer, still entirely in-perimeter.

Why does it sometimes say it doesn't know?

Because it will not guess beyond its corpus. A grounded system that admits the gap is far more valuable in a regulated setting than one that invents a confident wrong answer — hallucination is the single biggest blocker to enterprise AI adoption, and grounding with citations is how you defeat it.

Can I run this on my own documents?

Yes — that is the productised version. The same retrieval-grounding-citation pattern, deployed inside your perimeter on your policies, contracts or knowledge base, served by a sovereign model with your SSO, your audit trail and no data egress. Ask a couple of questions here, then leave a note and an engineer will scope it.

What product is this?

The conversational layer is MindMap ChatNext — our multilingual assistant for WhatsApp, web and voice — grounded on a sovereign RAG stack (pgvector or Qdrant, BGE-M3 embeddings, re-ranking) inside the customer's perimeter.