It works like ChatGPT or Claude, but it is ours: it runs on Uni-edit's own Microsoft Azure account, only named staff can sign in, and nothing you type goes to a public chatbot service.
If you have used ChatGPT or Claude, you already know how to use it. Type a question, read the answer, ask a follow-up.
Start with Logging in, then Your first hour. You do not need to learn anything technical. You type in ordinary English — or Chinese, or Japanese — and it answers.
Read What is different here. The short version: same conversation, same uploads, same model picker. Different in that it is private to Uni-edit, the models are chosen for our work, and there is no "Projects" feature — reusable assistants do that job instead.
Go straight to Working with documents. Upload a manuscript, a spreadsheet or a PDF and ask about it, compare two versions, extract details, or have it suggest changes.
Yes — Using it on customer work says what you are accountable for, and What you can put into it says where your text goes. Read both if you have seen an older version of this manual, which said the opposite.
| Write, rewrite, summarise, explain, translate your notes | Yes |
| Upload a document and ask questions about it | Yes |
| Choose which AI model answers | Yes — seven available |
| Search your past conversations | Yes |
| Save a reusable assistant for work you repeat | Yes |
| Look up Uni-edit customers, jobs or experts | Not yet — see below |
| Remember you between conversations | No |
| Search the web | No |
It has no access to our business records. Asking "show me this customer's job history" will not work — it cannot see the customer database at all. That connection is built and tested but not switched on, because who may see which records has to be right before it is turned on, not after.
You will be told when it changes, and what each person can see will match their own work.
What it cannot do yet has the full list, and what to do instead in each case.
Anything you send to a customer is from you, not from the assistant. It goes out with your name on it. The assistant can be fluent, well-organised and wrong — a drafted email can read beautifully and contain the wrong date, the wrong price or the wrong name.
Read what it writes the way you would read a draft from a capable new colleague: worth having, not yet trusted.
First, start a new conversation and repeat the request clearly — most oddities go away. If it continues, send Daniel:
Do not send passwords or anything confidential that is not needed to explain the problem.