Short answer: our work, including customer manuscripts. Not credentials, and not other people's confidential material that is not ours to share.
An earlier version of this system made you pick a category — public, customer metadata, customer deliverable, unknown — from a dropdown before every request. There is no such dropdown here. Nothing asks you to classify anything, so the judgement is yours, and this page is what you need to make it.
| What | Where it goes |
|---|---|
| The message you type | Uni-edit's own Microsoft Azure account, processed in the Asia-Pacific region |
| Files you upload | The same, plus our own server, so the assistant can search them |
| Your conversation history | Our own server in Hong Kong, backed up nightly to Uni-edit's Azure storage |
| Anything at all | Not to ChatGPT, Claude, Google or any public chatbot service |
Nothing you type is used to train any AI model. Your conversations are not visible to other staff.
Daniel can see usage totals — how many questions, which model, what it cost — because someone has to run the trial and pay for it. He does not read your conversations.
This is a real change from what you may have read before, and it was a deliberate decision rather than an oversight.
Every model here is deployed so that it processes your text within the Asia-Pacific region. That is what makes customer material acceptable to put in — not a promise, but how the service is configured. It is also why the model list is what it is: models that could not be pinned to this region are not offered, however good they are.
So a customer's manuscript, an author's draft, a job's details, an expert's notes — all fine.
Passwords, API keys, or anything you would type into a login box. Never, in any system, for the ordinary reason: a conversation is stored, backed up, and searchable, and a password in it is a password in all three. If you need to tell Daniel about a credential problem, describe the problem, not the credential.
Material that is not ours to share. A document another company gave us under a confidentiality agreement is governed by that agreement, not by this page. If you are unsure who owns something, that uncertainty is the answer — ask before pasting.
Anything where you would not be comfortable saying "I put this into Uni-edit's AI assistant." That sentence is a better test than any category list, because it is the sentence you would actually have to say.
Conversations are stored and searchable, which makes it tempting to treat the portal as the place a piece of work lives. Do not.
Keep it as the place you did the thinking, not the place the outcome is kept.
Worth stating plainly, because it is the whole reason this page replaced the old one: the old dropdown was not really a safety control either — it told the system how carefully to treat your material, and it could be set wrongly. What has changed is that there is no longer even that prompt to stop and think.
That stopping-and-thinking is now entirely yours, and it takes about two seconds. See Using it on customer work for what you are accountable for once the answer comes back.