The fastest way in is search. Press ⌘K / Ctrl+K (on a phone — the magnifier in the header) and type your question in your own words: the "Help" group lists matching pages of this guide, each opening in a new tab. Help search is available to every staff member.
The assistant (chat) is for the owner and the manager of a venue: a button at the bottom of the sidebar, an icon in the phone header, the "Assistant" row under "More", and the ⌘/ (Ctrl+/) hotkey. A shared device (a POS or kitchen station) has no assistant — there is nobody to hold a conversation with.
One assistant is open at a time: opening the menu helper in the menu editor closes the chat, and the conversation stays — open the chat again and it is there. A question that is not about the menu, asked of the menu helper, is passed here — the chat opens with your phrase.
A conversation belongs to one venue: switching to another venue shows that venue's own list of conversations.
Conversations split themselves: come back after a few hours and the next question opens a new one, while the previous stays under "History" — the list behind the conversation's name in the panel header. You never have to press "New conversation", though the row is there. The assistant only remembers the last few turns of a conversation, so one endless chat buys nothing: it re-reads the venue's settings and the guide for every question anyway.
The assistant answers only from this guide and from your venue's settings. Ask like you would ask a colleague: "why does the waiter have no floor screen", "how do I translate the menu into English", "where do I print table QR codes". Answers are short and link to the relevant pages.
It replies in the interface language. When the guide has no answer it says so and offers to write to support — it does not invent features, prices or limits.
Every answer has 👍/👎 underneath — the marks tell us where the guide fell short.
The microphone button next to the input field starts a conversation: tap once and just talk. The assistant works out that your phrase has ended, sends it, reads the answer aloud and listens again — no hands needed. Tap the button again to finish; if you stay quiet for half a minute the conversation closes itself and releases the microphone.
In Chrome and Edge the words appear in the chat while you are still speaking. Elsewhere (Firefox, Safari) speech is recognized on our server: the phrase is sent once you go quiet — a second slower, but it works everywhere. The audio is transcribed on the fly and stored nowhere. You can speak Russian, English or Vietnamese.
If the browser asks for microphone access, allow it — without access the button says so. The microphone stays open for as long as the conversation runs, so on a shared tablet in the dining room end it with the button rather than just closing the panel.
Separately from conversations, the "⋯" menu has a "Read replies aloud" switch: with it the assistant also reads answers to questions you typed. Off by default, and only on this device.
Ask "show me where the order mode is" or "where do I turn on the kitchen screen" — the assistant opens the screen and highlights the element with a short caption, like the onboarding tips. On a phone the assistant window folds into a small chip at the bottom while it shows; tap the chip to return to the chat.
The assistant never presses buttons for you: it shows, you act. If your role has no access to a screen, it tells you so instead of showing.
"Turn on full order mode", "set the geofence radius to 150 metres", "switch the waiter screen off" — the assistant proposes the change as a "was → will be" card. Nothing changes until you press Apply; risky changes (turning orders off, enforcing presence) ask for one more confirmation. An applied change can be undone the same way.
A request to change the menu — “raise the pho price”, “hide the rolls”, “fettuccine is sold out” — goes to the Menu helper: the assistant hands your phrase over as-is, the menu editor opens with an edit card, and you confirm it right there. No need to repeat yourself.
A proposal is valid for 15 minutes. If someone changes that setting by hand in the meantime, the card asks you to phrase the request again — so nobody's edit is overwritten.
The assistant only changes venue settings from "Settings": order and presence modes, staff screens, kitchen thresholds, ordering hours, pre-bill printing, currency and language, name and contacts. Plan and payment, deleting the venue, staff access, passwords, POS connections and collecting guests' names and room numbers it does not touch — it shows where that is done and points to the guide.
Each message carries only structure: which screen you are on, the interface language, phone or desktop, and which interface elements are on screen — by their ids, without text. The assistant reads the venue settings from the same list shown in "Settings", and the text of this guide.
It does not receive passwords, sign-in codes, POS integration keys, allowed address lists, e-mails and phones of staff or guests, payments, or data of other venues. Dish names, guest comments and the names guests gave on their orders never enter the conversation.
Conversations are kept for 90 days and are visible only to you; the Qreaty team sees the requests you send to support, so they can answer them.
The "⋯" menu of the assistant and the bottom of the search have "Need help" — a short form that attaches the context (screen, role, language) automatically, so "it does not work" is understood at once. The reply comes to the account's e-mail or to Zalo if you listed it for the venue.
The team can also answer right inside the conversation — a human reply appears in the same assistant window, marked as coming from support. When you are done, the "⋯" menu can close a conversation; it stays in the list read-only.
The same form appears when the assistant cannot answer (no connection to the model or the daily limit is used up): you see the three closest guide pages and the contact button.