QuorelloGet startedBlog › Privacy
Zero Data Retention, Explained for Professionals
If you work with client files — a fiduciary reviewing a tax position, a lawyer sketching an argument, a compliance officer checking a policy — the useful question is rarely just "is this AI any good?" It's "what happens to what I type in?" Zero Data Retention (ZDR) is the answer to that second question, and it's worth understanding plainly.
What "retention" actually means
When you send a prompt to an AI model, that text travels to the provider that runs the model. To generate a reply, the provider has to read your prompt. The open question is what happens afterwards.
By default, many providers keep the text for a while — to debug their systems, to review for abuse, or to improve future models. That's retention. Some also use submitted content, in aggregate, to train the next version of the model. Neither is inherently sinister, but both mean a copy of your input now lives on someone else's infrastructure, potentially readable by their staff and potentially baked into a future model's behaviour.
Zero Data Retention flips that default. A ZDR endpoint reads your prompt, produces the answer, and then does not retain the prompt. Concretely, that means:
- The provider does not keep your prompt after the response is returned.
- The provider does not train its models on your prompt.
For a professional, the difference is not abstract. If your input mentions a client's name, a salary, a disputed figure, or the shape of a case, ZDR is the line between "answered and discarded" and "answered and stored somewhere I can't see or control."
How this differs from a consumer chatbot
The free chatbot most people know is built for a different purpose. Its default settings typically retain your conversations and, unless you dig into a setting, may use them to improve the product. That's a reasonable trade for casual use. It is a poor fit for professional secrecy: you would not photocopy a client's file and leave it in a shared drawer, and pasting the same content into a consumer chatbot on default settings is closer to that than most people realise.
The distinction is about defaults and guarantees, not about which company is trustworthy. A consumer tool optimises for a smooth product loop. A ZDR endpoint optimises for not holding your input at all.
Quorello's Private mode
Quorello is a Swiss-built tool that puts your question to several AI models from independent providers at once. Because handling sensitive input is the point, Private mode is on by default. When it's active:
- Requests route only to ZDR endpoints — providers that don't retain your prompt and don't train on it.
- Models without a ZDR endpoint are disabled for that question, so nothing quietly slips through a non-ZDR path.
- The conversation is not saved to your history — there's no stored transcript on our side.
You keep the core benefit of the product — cross-checking several independent providers to see where they agree, diverge, and how confident to be — without your input being retained by the models that answer it.
The honest limit: a model still reads your prompt
Here's the part a lot of marketing skips, and we won't. A provider has to read your prompt to answer it. That's not a Quorello choice; it's how language models work. Because the model must process the actual text, your prompt content cannot be end-to-end encrypted the way a sealed message between two people can. ZDR means the provider doesn't retain your input — not that it never sees it.
So the practical rule stands:
- Don't paste anything you wouldn't be willing to share with a competent, discreet third party.
- Anonymise client data where you can — swap real names for placeholders, drop identifiers that aren't needed for the question. You almost always get the same quality of answer, with less exposure.
ZDR meaningfully shrinks your data footprint. It does not turn a prompt into a secret only you can read. Treating those two as the same thing is exactly the kind of overconfidence a verification tool should help you avoid.
What else Quorello does with your data
Beyond ZDR routing, a few commitments that matter for professional use:
- No selling of data, and no using your prompts to train AI models — full stop.
- Uploaded files are extracted in memory only — never written to disk, database, or logs.
- Self-service account deletion, so you can exercise your right to erasure (GDPR / FADP Art. 17) without emailing anyone.
- Built in Switzerland and aligned with FADP (Swiss) and GDPR principles.
None of this makes you compliant on its own — that stays your call as the professional who signs off — but it gives you a verification layer designed for sensitive questions rather than one built to harvest them.
In one line
ZDR means the models answer your question and then let it go. Quorello's Private mode makes that the default, disables anything that doesn't honour it, and keeps no transcript — while being honest that a prompt still has to be seen to be answered, so good judgement about what you paste still matters.