FlowsBook a callSign inCreate free accountSign up
AI employees and automation

An AI agent for restaurants: answer guests, and keep an eye on the street

Two AI employees a restaurant can run: a chat page that answers guests about the menu, hours and parking, and a weekly watch on new and changing rivals nearby.

Part of the guide AI employees

By the xMap team · · 4 min read

An AI agent for restaurants does two jobs well. Facing out, it answers guests on a chat page: is there a vegan option, are you open on Monday, where can I park, will the terrace be warm enough tonight. Facing in, it watches the streets around you each week and tells you which places opened and which rivals' ratings moved. In xMap Flows those are two separate AI employees: Chat, an AI agent and Reply for guests; and a Schedule with Nearby places, Only new and a written summary for you.

This is for owners and managers of one to a dozen sites who answer the same messages every evening and hear about a new opening only when a regular mentions it.

The guest chat

Guests ask about four things: the food, the hours, getting there and the booking. The agent can handle the first three if you give it the facts, and should hand the fourth to whatever you already use.

  • The menu and allergens: upload the current menu as a document and switch on Look things up in for the agent. It answers from the passages it finds. Ask it to say plainly when an allergen is not listed, rather than reassure.
  • Hours and branches: one row per site in a My data table, with weekday, weekend and holiday hours. Linked as a tool, the agent reads it when asked.
  • Parking and transport: Nearby places as a tool, searching car parks and stations near the site.
  • The terrace: Weather forecast as a tool, for tonight or Saturday.

For bookings, tell the agent to give your booking link or phone number, which you list as contact details it may share in the Protection settings. A chat agent that tries to take bookings with no system behind it creates no-shows.

The weekly watch on rivals

The inward job is simpler. A flow runs once a week:

  1. Schedule, Monday at 9, before the week's planning.
  2. My data with your sites, or an Area of interest drawn around them.
  3. Nearby places searching restaurants (and cafes, if they compete for your lunch) within a kilometre of each site.
  4. Only new, so last week's places are not reported again. In parallel, What changed on the same results flags rivals that closed or whose rating moved.
  5. xMap LLM on the whole list: for the owner, three lines per site: what opened, what closed, any rival whose rating fell or rose by half a star.
  6. Email to the owner and the head chef.

If you would rather hear about an opening the moment it is found than wait for Monday, the When something changes trigger checks around up to five of your sites on a timetable you pick and runs only when a new restaurant appears within the distance.

What the numbers look like in a busy city

To get a sense of the churn, here is one city in xMap's places database:

2,125Restaurants listed
22 of 290Opened in the last year
Restaurants within 10 km of the centre of Toronto, xMap's places database, November 2026; the openings are of the 290 nearest the centre that have an opening date.

Twenty-two openings among the 290 restaurants nearest the centre is a few a month in a small part of one city. A weekly list near your own door will often be short, and that is the point: two or three names you can go and look at.

A worked example (made up)

A family-run Lebanese restaurant in Toronto with two sites gets a stream of evening questions on Instagram and by phone about gluten-free dishes and parking. The owner's son builds the guest chat from the parking and terrace weather chat template, adds the menu with allergen notes as a document and the two sites as a table, and puts the chat page link in the restaurant's profile.

He sets up the weekly watch as well. In its third week it reports a new grill house 600 metres from the second site and a long-standing rival whose rating has dropped. The family tries the grill house that weekend and changes nothing; the rival's drop they note, and watch.

Mistakes to avoid

  • An outdated menu document. Replace the file whenever the menu changes, or the agent will describe last season's dishes.
  • Letting the agent give medical reassurance about allergies. Have it state what the menu says and suggest guests confirm with staff.
  • Watching too wide an area. A kilometre around each site in a city, a few kilometres in a town, is usually enough.
  • Reading a falling rating as a fact about quality. It may be a few angry reviews; it is a reason to look, not a verdict.

Both employees are covered in more depth in the AI employees guide, and the new restaurants opening near my branches template sets up the watch in a few clicks.

Questions

Can the chat page take reservations?

Not on its own. Have it give your booking link or phone number, or connect the system you already use for bookings.

Will it make up dishes that are not on the menu?

Tell it to answer only from the menu document and to say when something is not listed. Check the work can also reject answers that mention items not in its data.

How often should the rival watch run?

Weekly suits most restaurants. A daily run tends to report nothing most days.

Can my staff use the chat too?

Yes. The same page works for staff questions about sites and hours, and the editor's Chat lets you try it without sending anything.

Templates that do this

Read next

More on ai employees and automation · Every article · Templates

Put it to work

An AI employee in xMap Flows does this on live data, on its own timetable. Start from a template or describe it in plain words.