The question lands in almost every first sales call, and it is fair: "Can I just connect my ChatGPT — or my Claude — to my customer chat?"
The short answer is yes. The important answer is: not the way you're picturing it. Wiring a raw frontier model straight into a website chat is exactly the liability machine Wegenty exists to prevent. It has no governed knowledge base, no citations, no grounding check, and no honest way to say "I don't know." A confident, uncited, invented answer to your customer is a real-world risk — and legal uncertainty is already the single most-cited reason SMEs hesitate to adopt AI at all.
So we flip the framing. You don't connect a model to your customers. You connect a model to Wegenty's governance layer, and Wegenty speaks to your customers. The model becomes one configurable component. Everything that makes the answer trustworthy stays exactly where it is.
The model is the engine, not the car
Here is the uncomfortable truth for the "powered by [big AI name]" pitch: the model is a commodity. A swappable engine. It is not the product.
The product is the governed pipeline around the model — the governed projection of your knowledge, the grounded retrieval, the citation check, the escalation path. That pipeline is the same whether the engine computing the tokens is a small model running on your own box or a frontier model in a vendor cloud.
Swapping the engine changes who computes the tokens, not what the agent is allowed to say.
That distinction is the whole point of this post. Because inference already runs behind a single internal facade, you get to choose which brain powers your public chat — without touching a single governance guarantee. Model choice is a performance, cost, and sovereignty dial. It is not a governance dial.
Three tiers of brain
We don't push customers up a ladder of ever-more-powerful models. We make the safest option effortless and the rest an informed, logged decision your consultant makes with eyes open. Three tiers:
- Local (default) — full sovereignty. A model runs on your own box. No question text ever leaves the premises. This is the Cloud Shield promise in its purest form: DSGVO-clean, predictable flat cost, nothing to leak because nothing leaves. It is the default for a reason.
- Frontier-EU — partial sovereignty, on request. Claude or GPT, pinned to an EU region (Frankfurt, Ireland, Paris) with EU data residency and no training on your data. Your data stays in-EU but does leave the box. The right choice when a customer genuinely needs frontier-grade reasoning and accepts a processor relationship with an EU-resident vendor.
- Frontier-global — opt-in only. The default Claude or ChatGPT endpoints, US-default cloud. We flag this honestly as breaking the sovereign default. It exists for the rare non-sensitive case, and it is never the recommended path.
The job of this menu is not to upsell. It is to make local the effortless default and the cloud tiers a deliberate, per-instance choice — sovereign by default, not by upsell.
The invariant nobody gets to break
Whichever brain you pick, the things that make the answer safe do not move:
- The public agent still only sees the governed projection of your knowledge — never the raw internal base.
- Every claim is still cite-or-silent: attributed to a real chunk, and when grounding confidence drops below the 0.85 threshold, the agent retrieves again or says "not found" rather than guess.
- A grounding judge still discards any draft the evidence doesn't support, structurally.
- An unsure agent still escalates to a human instead of bluffing — and that human's answer flows back into the correction loop so the system compounds over time.
None of that depends on the model. A bigger engine writes nicer prose; it does not earn the right to invent. The governance layer enforces the same contract on a frontier model that it enforces on a 7-billion-parameter local one. That is why "which model" is a safe question to leave open — and a dangerous one to answer without the layer underneath.
The honest sales line
We can say this plainly, and we do:
"Bring whatever brain you trust. We make it cite its sources, stay inside your approved knowledge, and hand off to a human when it's unsure — and we do that the same way whether the brain is on your box or in the cloud."
That sentence is the entire pitch. It is also the reason we stay engine-agnostic in our own marketing. "Powered by Claude" or "powered by ChatGPT" is a tempting badge, but it points the spotlight at the part that is replaceable. The moat is the governance, not the model. The model is just the part you're free to change your mind about.
Where this leaves you
If you already trust a model — your Claude, your ChatGPT, or a local one you run yourself — you don't have to give it up to adopt Wegenty. You connect it to the layer that makes it safe to put in front of a customer, and you keep the freedom to swap it later when something better ships. The platform page walks through the full pipeline the model plugs into, and the comparison page shows how this differs from bolting a chatbot onto your docs.
Want to see a governed agent answer on real knowledge — cite where it can, escalate where it can't? Try the live demo. When you're ready to talk about which brain fits your business, get in touch and we'll scope it with you.