Use of artificial intelligence

Last updated: 17 September 2026

There are three places in Qder.ai where a language model answers: the digital waiter at the table, the "busy right now?" question on the home page, and the assistant in the business panels. This page explains exactly what each of them receives, what it will not do, and when it stops trying and hands over to a person.

This English text is provided for convenience. The Hebrew version of this document is the one that governs: שימוש בבינה מלאכותית.

1. Who is answering you

You are talking to a system, not a person. We do not pretend otherwise: if you ask the digital waiter who it is, it will say it is Qder.ai. Behind the scenes we use OpenAI's GPT-series models through their commercial API. Under OpenAI's commercial terms, what is sent through the API is not used to train their models. We do not train models on your conversations. What we do use: a scrubbed log of conversations, to improve the waiter at the restaurant where you were talking, and only there. See section 11.

2. The digital waiter at the table: what is sent to the model

Sent:

• the message you wrote, and the most recent messages in the current conversation (kept on the server for 24 hours, see the privacy policy);
• the menu of the branch you are sitting in, with prices and the allergen list as the restaurant entered them, and what has run out;
• what has already been ordered at your table, including an allergy note if you wrote one, and how you asked for each dish to be prepared;
• the questions the kitchen asks about each dish before preparing it — how well done, how spicy, extras — as the restaurant defined them, or as the system inferred from the dish name where the restaurant did not (grilled meat asks how well done). The digital waiter asks you a required question before it adds a dish, and passes your answer to the kitchen with it. An answer that is not on the restaurant's list is not accepted;
• the allergies and dietary preferences saved in your account, if you have an account and filled them in;
• your previous visit to that chain (what you ordered and when), if you have an account, so it can answer "what did I order last time?";
• the fact that you are under 18, if the birth date in your account says so, so that it neither suggests nor adds alcoholic drinks. The date itself is not sent;
• how long this kitchen is taking to serve, based on its recent orders, so it can answer "how much longer?".

Not sent: your name, email address, phone number, birth date, IP address, or visits beyond the most recent one.

Allergies are sent so that the answer is about you rather than about an average diner: when you ask "what do you recommend?", the digital waiter should know not to suggest something you are allergic to. Anyone ordering without an account, or who has not filled those fields in, sends none of this. You can delete them at any moment on the account page, and from that moment they stop being sent.

3. A great deal of what you ask never reaches the model at all

"Menu", "how much is X", "what is there to drink", "call a waiter", "the bill" — all of these are answered directly from the restaurant's own data, without going to the model. It is faster, cheaper, and above all more accurate, because filtering our allergen list cannot invent a dish that does not exist.

4. What the model will not do

• It will not give medical advice, and it will not confirm that a dish is safe for you. With an allergy, or in any doubt, it calls a waiter.
• It will not answer things unrelated to the meal. This is not a system for general questions.
• It will not invent dishes, prices or ingredients that are not in the menu the restaurant entered — and even if it fails at that, what binds is the restaurant's menu, not what was written in the chat.
• It will not send an order to the kitchen without you confirming it on screen.

5. "Busy right now?" on the home page

On the home page you can ask, without sitting at a table, whether a particular restaurant is open and busy. That conversation is open only to people signed in to an account. The answer is built from the opening hours the restaurant entered, the number of tables occupied right now, and how busy it has been in recent weeks. Only your question and that picture are sent to the model — not your name, email or phone number. The system does not list which restaurants work with it unless you asked.

This is an estimate, not a promise: a free table is generally not a table held for you. For that there is a reservation.

6. The assistant for businesses

The branch and chain panels have an assistant that takes instructions in plain language — "the salad has run out", "the cola is 14", "here is a photo of the menu, type it in" — and proposes changes. A few rules that cannot be bypassed:

• No change takes effect without an employee's explicit approval. The assistant proposes; a person presses "confirm". Every proposal is signed by the server and expires if it is not approved.
• The assistant can only do what the account it was launched from is allowed to do. A branch manager cannot, through the assistant, change the chain's menu.
• Messages, images and files sent to the assistant are passed to OpenAI for processing. Do not send the assistant diners' details, financial documents or passwords — it does not need them.
• A menu imported from a photograph is a proposal from a model reading an image. The business approves it, and the business is responsible for it — prices, names and above all allergens.

7. Keeping answers

So as not to ask the model the same question a hundred times a day, we keep answers to general questions at branch level. Three important limits:

• No answer touching ingredients, allergens or nutrition is ever kept. That is exactly where a stale answer is dangerous, so those are answered afresh every time.
• For a diner with allergies saved in their account, or whose account says they are under 18, no answer is kept and none is read. Their answers were written for them, and they are not right for anyone else — neither in that direction nor the other.
• Any change to the menu — a price, a dish that has run out, a new dish — clears every kept answer for that branch.

8. Limits of use

Every diner has a ceiling on conversations with the model, and every chain has a monthly quota. When they are reached, the digital waiter stops answering, while ordering from the menu, calling a waiter and the bill keep working as usual — so that nobody is left stranded in the middle of a meal. The system also calls a waiter at that moment.

9. No automated decisions about you

The system does not score you, does not make decisions with legal significance about you, and does not build a behavioural profile. It answers questions and passes orders to the kitchen.

10. Mistakes

If the digital waiter said something wrong, we would like to know — here.