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

AI and tools

San and MCP: one memory, many ways to talk

Procedures, tasks, logs and incidents stay in one place. San by voice, Tsuno on screen or your AI assistant: same permissions, same rules.

Published on

In the evening, a restaurant manager asks her laptop a question in the empty dining room while, in the kitchen, a cook finishes putting things away.

In Tsuno, your business's facts live in one place: procedures, tasks, checklists, temperature logs, incidents. There are three ways in: by voice at the workstation with San, on screen in Tsuno, or from your usual AI assistant at the office. From one door to the next, permissions, confirmation, refusals and the log stay put, so there is only one version of the facts and nobody types anything twice.

Tuesday, 10:15 pm, end of service

Lise is a line cook. She is finishing the dishes, hands in the water. She asks quietly: "What's left on the closing checklist?" San answers in one sentence: check the walk-in, label yesterday's containers, log the prep fridge temperature. She adds: "The dishwasher is leaking, on the left side." San prepares the incident report, reads back a summary and waits. Lise says yes. It is recorded, under her name.

At 10:50 pm the dining room is empty. Gaëlle, the general manager, opens her laptop on a table at the back. In the kitchen, Antoine is finishing the clean-down. Gaëlle does not feel like opening five screens. She types to the AI assistant she uses all day: "Is tonight's closing complete? Were there any incidents?"

The answer comes back: the checklist is done except the prep fridge log, and an incident was reported at 10:16 pm by Lise, dishwasher leak. Gaëlle asks for a task for the morning team: call the technician before 10 am. The assistant drafts the task and summarizes it. She confirms in a new message. At 7 am the next day, Mehdi finds it on his list.

Nobody copied anything. Lise, Gaëlle and Mehdi used three different doors into the same memory.

(Fictional scene and characters.)

Why three doors beat three tools

The real problem is the second version of the truth

In many restaurants, the same information exists in several copies. The shift logbook says one thing, the group chat another, the office spreadsheet a third. At night, the manager does not know which one to trust. Each copy was written in good faith, but they never line up.

An AI assistant fed with a separate document would only add a fourth copy, the very one to avoid.

Retyping is where mistakes get in

Every time a fact moves from one medium to another, it can be distorted, forgotten or misdated. A temperature written on a scrap of paper and copied in the next day is no longer quite the same reading. The less you retype, the fewer mistakes you make.

Everyone has their own best door

The line cook has wet hands: voice suits her. The sous chef needs to see the whole prep list at a glance: the screen is better. The manager spends her evening on her laptop with her assistant already open: she would rather type to it. Forcing one door on everyone means losing part of the team.

What does MCP change for you?

Behind the acronym (Model Context Protocol) is a simple idea: the assistant you already use comes to Tsuno for the information, instead of you copying your data into the assistant.

The acronym hardly matters. For you, it mostly changes this:

  1. Your facts do not leave Tsuno to be copied elsewhere. The assistant asks a question, Tsuno answers with what it has.
  2. The assistant works with your permissions, not with general access.
  3. Actions follow the same path as in Tsuno: draft, summary, confirmation, log.

We explain the idea in detail, without jargon, in MCP explained for frontline trades.

Which rules apply, whatever the channel?

Changing doors does not change the rules. Here is what stays the same, whether you talk to San, tap in Tsuno or type to your assistant.

  1. The person's permissions. No channel widens the scope: Lise cannot use her assistant or San to look at a colleague's schedule she does not normally see. More in give your AI the same rights as you, no more.
  2. Confirmation. By voice, Lise says yes after hearing the summary; in writing, Gaëlle replies in a new message after reading it. That is the two-step rule, the same everywhere.
  3. Permissions checked again at confirmation. If access was removed between the draft and the yes, the action does not go through.
  4. Refusals. Tsuno does not rate a person in place of their manager, and refuses to record an employee's health, pregnancy, disability, religion or opinions (sensitive data under Article 9 of the EU GDPR). It offers to log a dated, observed fact instead.
  5. What stays out of the AI's reach. Payroll, invoicing, permanent deletion and legal acts do not go through the assistant, even when confirmed.
  6. The log. Every confirmed action is recorded with who, when and what, along with the state before and after.

There is one more rule specific to office assistants: reads are capped, so the connection cannot be used to pull out all your data at once.

What ties these rules together fits in one line: Tsuno brings the facts together, people decide.

Voice, screen, assistant: which does what best?

All three doors open onto the same memory, but they suit different moments.

San by voiceTsuno on screenYour AI assistant
Where, whenAt the station, hands fullIn the kitchen, at the pass, in the officeIn the office, at night, on the road
Mostly forLine cooks, dish crew, opening, closingThe whole team, managersOwners, managers
Kind of questionShort: next task, next stepRead, compare, approveOverview: what happened, what is still open
PermissionsThe person's Tsuno permissionsThe person's Tsuno permissionsThe person's Tsuno permissions
ActionsDrafted, then confirmedDrafted, then confirmedDrafted, confirmed in a new message
RecordTsuno logTsuno logTsuno log
LimitsNot front of house, little use in the rushYou have to stop and lookNot built for the workstation
StatusSan Voice available in demoAvailableAvailable

On the voice side, San Voice, the voice interaction, is available in demo. San Audio, the hands-free experience, and San 360, a broader vision of operational support, are under exploration and testing. We do not present them any other way.

Common mistakes

  • Pasting your procedures into an AI assistant. You create a second version that will never be updated alongside the real one. The day the procedure changes, the assistant answers with the old one.
  • Giving the assistant a shared login or an admin account "so it works". The assistant should carry the permissions of the person talking to it, no more.
  • Treating a typed request as a confirmation. Asking "add a task" is not approving the task. Read the summary before you say yes.
  • Forgetting the assistant's own terms. What you type into your AI assistant is also covered by that assistant's terms. For work use, check your account settings.
  • Putting voice everywhere. Voice belongs at the station, in the kitchen and the stockroom. Front of house, in front of guests, it has no place. See why talk to your tools when your hands are full.

What to measure

  • How many times a piece of information was copied from one medium to another over a week (logbook, group chat, spreadsheet). You want that number to fall.
  • Gaps between what the floor says and what management sees, noted at the Monday meeting. If gaps remain, one of the doors is not connected to the same memory.
  • The share of actions drafted by an assistant and then edited or rejected before confirmation. It shows the review step is doing its job.

Where Tsuno and San fit in

Tsuno brings together your business's procedures, tasks, checklists, logs and daily logbook. San is the door designed for the workstation, by voice or on screen: you ask for your next task or a cleaning step, and it answers from what the team has written. From the office, you can query Tsuno with ChatGPT, Claude or Mistral, with exactly the same permissions as in Tsuno and the same confirmation before any action. The rules are public on the Trust and data page, and the list of connections is on the Integrations page. The Tsuno San page presents the assistant, and a demo lets you see the same question asked through two doors.

Further reading

For the idea of plugging an AI into your business, start with MCP explained for frontline trades, then read AI proposes, a human confirms: the two-step rule. For San as a whole, see the pillar page Tsuno San: the work companion for frontline teams and its red lines in what San never does. And if you wonder why not build it all yourself with consumer tools, read why not build Tsuno with ChatGPT, Notion and Make.

Key takeaways

Your business's facts live in one place, and everyone reaches them through the door that suits them: voice at the station, the screen, or their AI assistant at the office. Changing doors grants no extra rights and skips no confirmation.

Frequently asked questions

Can I query Tsuno from ChatGPT, Claude or Mistral?

Yes. Tsuno can be connected to the AI assistant you already use. The assistant only sees and does what you can already see and do in Tsuno, and every action waits for your confirmation.

Do I need to understand MCP to query Tsuno from my assistant?

No. It is the technical plug that links the assistant to Tsuno. For a restaurant operator, the real questions lie elsewhere: which permissions, which actions, which record.

Is the answer San gives by voice the same as my AI assistant's answer?

It comes from the same facts, the ones your business has recorded in Tsuno. The wording can vary from one channel to another, but not the source, and not the permissions of the person asking.

Can an employee see more by going through their AI assistant?

No. Switching channels gives no extra rights. A line cook who queries Tsuno from an assistant only sees what they already see in Tsuno.

Are actions taken from an AI assistant recorded?

Yes. The assistant prepares the action, you read the summary and you confirm. The action is then written to the log with who, when and what, exactly as if it had been done in Tsuno.

Are San Audio and San 360 available?

Not yet. San Voice, the voice interaction, is available in demo. San Audio and San 360 are under exploration and testing.