Meet Tsuno at Sirha Lyon 2027 · Stand S.57Learn more
EN FR
Try the demo
Menu

AI and tools

Asking your restaurant questions from ChatGPT, Claude or Mistral

Connected to Tsuno, your usual AI assistant answers from your own location's data, within your rights, shows its sources, and runs nothing without your yes.

Published on

In his parked car, a restaurant owner asks his phone a question before walking into his restaurant.

If you already use ChatGPT, Claude or Mistral every day, you can ask it about your restaurant, as long as the assistant is connected to your data. Connected to Tsuno, it answers from your location's own facts, within your rights and no further. A good answer shows its sources, flags what is missing, and any action it suggests waits for your confirmation.

Wednesday, 9:50 a.m., in the parking lot

Cédric runs two restaurants. This morning he pulls up at the first one, parks and turns off the engine. He has ten minutes before he walks in and gets swallowed by the delivery and the team's questions.

He opens the AI assistant he already uses for emails and quotes, and types: "What needs my attention today?"

A few months ago, the answer would have been polite and empty: some general advice on running a restaurant. This time the assistant is connected to Tsuno. It answers in four lines: a temperature reading over the limit last night on a walk-in cooler, the closing checklist not done at the second site, an overdue task for cleaning the hood. And one sentence that matters: "I have no weekly report for the second site this week."

Cédric walks in knowing where to start.

(Fictional scene and character.)

Why use the assistant you already have?

Habit beats features

Many owners have picked up an AI assistant to write, summarize or prepare a meeting. They open it ten times a day. Asking them to open yet another app for a simple question adds a detour, one you avoid by letting them type where they already are.

A question beats hunting for a screen

A dashboard answers the questions its designer thought of. In the morning, the real question is often broader: "Where should I go?", "What hasn't been handled?". Asked the way you would ask a deputy, it is faster than opening three screens one after another.

Without data, the assistant guesses

A general assistant with no link to your restaurant knows nothing about your coolers, your checklists or your team. It answers anyway, with well-phrased generalities. That is the real danger: an answer that sounds useful and rests on nothing. We go into this in why a general AI does not know your restaurant.

What changes once the assistant is connected to your data

Connecting an assistant to Tsuno does not hand it the keys to the building: it comes in through a door where Tsuno's usual rules apply.

  1. It inherits your scope. A site manager only sees their site; a supervisor only the sites assigned to them (we cover this principle in give your AI the same rights as you, no more).
  2. It relies on dated facts. A Tsuno answer can show the facts and sources it used: a logged reading, a completed checklist, a note recorded by a manager.
  3. It says when data is missing. Nothing recorded means "no data", not "all good", and the answer has to say so.
  4. It waits for your go-ahead. A suggested action arrives as a summary, and your next message is what sets it off, or not.
  5. It does not empty the stockroom. Reads are capped, so the connection cannot be used to pull out your data wholesale.

This way of plugging an assistant into a company's tools has a name you will hear more and more: MCP. We explain it in plain words in MCP: connecting your AI to your business, explained for frontline trades.

Ten useful questions to ask in the morning

Here are questions an owner or manager can ask from their usual assistant. The answer is only worth something if the facts exist in Tsuno and you are allowed to see them.

QuestionWhat the answer should containWhat it should not do
What needs my attention today?Three or four points, each with the fact and its dateDump a twenty-line list with no priorities
Were the opening checklists done this morning?Site by site: done, not done, or no dataConfuse "not done" with "not recorded"
Was any temperature reading over the limit this week?The equipment, the value, the date, the action loggedConclude there is no risk
Which tasks are overdue at the second site?The task, the due date, the person or station assignedPass judgment on the person who is late
What have my managers not dealt with since Monday?The open items, dated, with where they came fromGuess why they have not done it
How did the week go at the downtown location?The highlights from the weekly report, if there is oneMake up a summary when the report is missing
Were there any discrepancies on deliveries?The supplier, the product, the gap flaggedBlame a supplier without the fact
Who has not yet completed the hygiene training?Training status: approved or still to approveRank or score employees
Which reviews with staff are overdue?The planned reviews and their datesOffer an opinion on the person
Prepare a task to have the walk-in door seal checkedA summary: what, who, where, whenRun it without your confirmation

How to check an answer before acting on it

A clear answer is not always a right one. Before you act on what the assistant says, take thirty seconds to check five things.

  1. The source. Every important point should lead to a fact you can open: a reading, a checklist, a dated note. If nothing is cited, ask "what are you basing that on?".
  2. The date. A fact from Monday does not weigh the same as one from this morning. An answer with no date cannot tell you whether it still holds.
  3. The gaps. A good answer says "I don't have that information" or "no data for this site". Tsuno also flags data that is missing, contradictory or too thin to call it a pattern.
  4. The kind of fact. A logged fact, something a person declared, an attached proof and an AI inference are not equal. An inference has to be labeled as one.
  5. The action. If the assistant offers to create, change or send something, read the summary before you confirm. Your yes carries as much weight as a tap in Tsuno.

A simple test at the start: ask a question you know has no answer. If the assistant makes something up, stop trusting it on that topic. We explain why in why an assistant must be able to say "I don't know".

What Tsuno says publicly about this connection

On its Trust and data page, Tsuno states that you can connect it to Claude, ChatGPT or another compatible assistant, and ask your questions where you already work. Three rules frame that connection: it follows your rights, actions happen in two steps, and reads are capped.

The Integrations page puts it in one sentence: ChatGPT, Claude or Mistral can query Tsuno, they see what you see, nothing more, and every action waits for your yes.

For an owner, the same page adds a few limits worth knowing:

  • rights are checked again at the moment you confirm, so an action does not go through if access was removed in the meantime;
  • payroll, invoicing, permanent deletion and legal acts stay out of the assistant's reach, even when confirmed;
  • no ranking of people is passed to the assistant, and the instructions Tsuno gives it ask for dated facts, never a judgment about someone.

One point stays on your side: your usual assistant runs on your own account with its provider. Check its privacy settings, as you would for any tool.

Common mistakes

  • Asking an assistant that is not connected and believing the answer. Without a connection, it talks about restaurants in general, not yours.
  • Pasting exports into the chat to "give it context". You lose rights, dates and the audit trail, and you take data out of your workspace.
  • Confirming an action without reading the summary. Confirmation only protects you if you actually take the time to read.
  • Asking for an opinion about a person. Tsuno lines up facts; it does not score an employee in their manager's place. See AI should not judge your employees.
  • Sharing one assistant account between several people. The connection follows one person's rights: everyone should use their own.

What to measure

  • The number of calls and chase-up messages sent before 11 a.m. over one week, before and after.
  • Out of ten answers, how many show their facts and dates, and how many flag missing data when there is some.
  • The time between a problem appearing (a reading over the limit, a forgotten checklist) and the moment you see it.

Where Tsuno and San fit in

Tsuno keeps your location's memory: checklists, temperature logs, tasks, reports, training, dated facts. You can query it from Tsuno or from your usual AI assistant, with the same rights and the same confirmation rule. Inside Tsuno, that assistant is called San, and teams can also talk to it by voice (San Voice is available as a demo). Whether you go through San or through ChatGPT, Claude or Mistral, the memory is the same: we explain how in San and MCP: one memory, many ways to talk. To see this connection on a real case, the Tsuno San page introduces the assistant, and the demo lets you ask your own questions on made-up data.

Further reading

To decide whether a conversation with your data is worth it, start with talking to your restaurant: gimmick or new interface?. For the start-of-week routine, see Monday at 8 a.m.: what a multi-site operator really needs to see. And for the assistant as a whole, the pillar page Tsuno San: the work companion for frontline teams.

Key takeaways

Your usual assistant becomes useful for your restaurant once it is connected to your data, within your rights. Judge it on its sources and on its honesty when data is missing, and check that it waits for your go-ahead before acting.

Frequently asked questions

Can I ask ChatGPT, Claude or Mistral questions about my restaurant?

Yes, if the assistant is connected to your location's data. With Tsuno, you can query it from the AI assistant you already use: it sees what you see in Tsuno, nothing more.

Once connected to Tsuno, can my AI assistant see all of the company's data?

No. The connection follows your rights: the assistant only sees and does what you can already see and do in Tsuno. Reads are also capped so the connection cannot be used to pull out data wholesale.

Can the assistant create a task or change a schedule for me?

It can prepare it. It shows you a summary, and nothing runs until you have confirmed in a new message. The action is then logged.

How do I know whether the assistant's answer is reliable?

It should show which facts it relies on, with their dates, and say clearly when data is missing or contradictory. A smooth answer with no checkable facts behind it should worry you.

Are there things the assistant can never do in Tsuno?

Yes. Payroll, invoicing, permanent deletion and legal acts stay out of the assistant's reach, even if you confirm. It is not given any ranking of people either.