AI agent settings

How Yaplet is named and switched on, where it answers, how it writes, what it does when it doesn't know, and when it hands over to a person.

Where this is

Brand → (your brand) → AI agent. The page has a side menu: General, Answers, Human hand-off, Phone, Starters, Suggestions, Reports and Danger zone. This article covers General, Answers, Human hand-off and Danger zone — the settings of this one AI agent and nothing else: what it is called, whether it is switched on, where it answers, how it writes, what it does when it cannot answer, and when it passes the visitor to a person. Phone, Starters, Suggestions and Reports have articles of their own. One Save changes button in the title bar (it appears, with Unsaved changes and Discard, once you change something) covers General, Answers and Human hand-off. The permission is AI agent.

At the top of every section, a row of numbers shows how Yaplet has answered over the last 30 days: Questions, Answer rate and Handed to a person. You see it if you also have the Overview permission.

The instructions and the test chat are on the Playground. The text Yaplet follows in every answer, and a test chat that answers with it before you save, are on the brand's Playground tab, described in Playground and AI setup. It is the same text as Answer instructions below.

General

  • Name — what Yaplet is called. Visitors see it in the chat widget, and your team sees it in the inbox. Only Enterprise plans can change it; on every other plan the field is locked on "Yaplet" and says "Renaming Yaplet is available on Enterprise plans." A name can be up to 40 characters, and the name of another company's AI is refused. Left empty, it is Yaplet.
  • Enable Yaplet — the on/off switch; the badge at the top of the section reads Live or Off. While it is off, Yaplet does not respond to visitors at all.
  • Answers on — the channels where Yaplet answers customers by itself: Website chat, Facebook Messenger, Instagram and Email, one switch each. On a channel you switch off, every message goes straight to your team's queue and notifies them, exactly as if the brand had no AI agent — no AI replies, and in the website chat no conversation starters either. Workflows keep running there. Phone is in the list as a row without a switch: it shows how many phone agents the brand has and how many are live (or Set up a phone agent when there are none), and opens the Phone section, where each phone agent can be paused or resumed. While Enable Yaplet is off, Yaplet answers on none of these channels, and a line under the list says so; the switches keep their values for when you turn it back on. Every agent starts with all four on.
  • AI models — the models behind Yaplet's answers, shown for information: Answers and Quick tasks. The answer model is changed under Answers.
Switching a channel off has a few knock-on effects. On Messenger and Instagram, the tappable opening questions disappear from that platform. On email, the AI answer filter on Brand → Brand settings → Email has nothing left to filter, and that tab says so. A workflow's Send to the AI agent step hands the conversation to your team on that channel instead of asking the AI. And a Block from agents only ban on a conversation there is applied as a full block, because there is no AI to keep the visitor company. A visitor who has the chat open when you flip a switch simply has their next message go to your team.

Answers

Three groups you open and close — Writing style, AI model and How visitors get identified — each showing its current value while closed.

  • Tone — Friendly, which adds an emoji now and then to keep the chat light, or Formal, which gives the same answers and never uses emojis.
  • Answer instructions — extra instructions attached to every AI prompt. Use it for tone, house rules, or facts the agent should always have to hand, for example: "Always greet users by name. Never discuss competitor products. Our return policy is 30 days." The same text is edited as Instructions on the Playground, where you can try it in a test chat before saving.
  • AI model — the model behind all AI in your workspace, set in Answer model. It is the same setting as Settings → AI Models, so it also drives inbox drafts, newsletters and more. Where your plan allows choosing, you pick it here; otherwise it shows locked, with Upgrade for more advanced models and a See plans button.
  • How visitors get identified — what the agent tells a visitor who is not identified when they ask for an API tool or a workflow switched to Only for identified visitors. Write how someone signs in on your own site and gets back to the chat, in the field Message to unidentified visitors. Plain text, at most 1,000 characters. The agent passes it on in its own words, in the visitor's language. Left empty, visitors are told the standard sentence shown in the box: "To use this, please sign in to your account on our website and open the chat from there, then ask again."
This text appears wherever the switch is. Beside the Only for identified visitors switch on an API tool and on a workflow's Start node, Yaplet shows what visitors will be told, with an Edit this text link back to this field. Who counts as identified is explained on Identity Verification.

Human hand-off

This section gathers every way a visitor can reach a person from your team in this chat, in the order the visitor meets them, and the check that lets the AI try first. It has two groups: Ways to ask for a person and When Yaplet has no answer. The AI never offers a person on its own: a person joins through one of these paths, or through a workflow with a Request agent step or an API tool set to hand over, which work exactly as before. What the visitor sees in the chat window is described on Email, Ratings & Agent Handoff.

  • “Request agent” in the chat menu — the visitor can ask for a person at any time from the menu (⋮) at the top of the chat.
  • “Talk to support team” button — a button under the conversation once the visitor has been chatting for a while. The line under the switch, Show it after the visitor has sent … messages in the last 5 minutes, sets how long that is: 6 by default, at least 1. Switched off, the button never appears. Until you save this switch for the first time, it follows Visitor asks Yaplet for a person: on when that switch is on, off when it is off.
  • Visitor asks Yaplet for a person — when the visitor tells the AI they want a person, the AI connects them with your team: it asks whether they want to be connected and hands over when they say yes. If nobody from your team is available, it asks instead whether they would like to leave a message for the team, and a yes opens a card where they write it. With Let Yaplet try first on, a check comes first (see below). On Messenger, Instagram and email, which have no buttons, the AI hands over right away.
  • Let Yaplet try first — off unless you switch it on; explained below.
  • When Yaplet has no answer — its own group, below the others: what happens when the AI finds nothing to answer from. The options are explained below.

The line the visitor reads immediately before being connected is one of Yaplet's own sentences, said in whatever language the visitor is reading, so a German visitor gets it in German without you writing anything. See Widget Languages.

With 'Visitor asks Yaplet for a person' off, Yaplet connects nobody. When a visitor asks for a person, it points them to Request agent in the ⋮ menu if that item is switched on. Otherwise — and always on Messenger, Instagram and email — it shares the contact details it finds in this brand's knowledge. To handle these requests your own way, build a workflow on this brand's Workflows page, set its trigger to User says, and give it the intent "The user wants to talk to the support team". The screen offers a link straight there whenever the switch is off.
"Visitors have no way to reach a person in this chat." This warning, with an Open this brand's workflows link, appears at the top of the section when the first three switches are off, When Yaplet has no answer is set to Politely decline, and none of this brand's switched-on workflows contains a Request agent step, counting workflows reached through a Start a workflow step. When such workflows exist, the tab names them under the switches, as Workflows that hand over: …, because they reach a person whatever the switches say.

Team hours

At the bottom of the section, a gray line shows the brand's team hours — when your human agents are usually online — as a short summary such as "Team hours: Mon–Fri 09:00–17:00, Europe/Budapest time." If none are set, it says so: "Team hours aren't set, so Yaplet says nothing about when your team is online." The line is read-only; Open team hours takes you to Brand → Brand settings → Team hours, where you change them, because they describe the whole brand, not this one agent.

With team hours set, Yaplet tells visitors when they can reach a person — but only when a visitor asks when the team is available, whether anyone is there, or asks for a person. It always says the team is usually there, never promises it, and gives times in the visitor's own time zone when it is known. A question like "When is your support team usually available?" is answered, not treated as a request to be connected. These are your team's hours, not your opening hours — Yaplet never uses them to say when your business or shop is open. Details: Brand settings.

When Yaplet has no answer

Choose what happens when the agent finds nothing relevant in this brand's knowledge:

OptionWhat happens
Politely declineThe agent tells the visitor it has no information on that topic. It does not guess and does not invent an answer. If the visitor asked it to do something — cancel an order, issue a refund, change an account — it says plainly that it cannot do that itself.
Connect to a personThe visitor is handed to your team right away, with no Yes or No question first. If nobody from your team is available, the agent tells the visitor so, and in the website chat they get a card for leaving a message for the team.
Connect to a person can only be chosen while Visitor asks Yaplet for a person is on. With that switch off, the option is grayed out and the setting falls back to Politely decline.
This applies when nothing at all was found. When a message asks about several things and your knowledge covers only some of them, the agent answers those and says plainly it has no information on the rest — it does not decline the whole message or connect a person for it. How the lookup works, topic by topic, is on How Yaplet finds answers.

When it declines, the agent writes a natural sentence in the visitor's own language rather than sending a fixed template. It mentions a way to reach your team — your website, a support e-mail address — only if it was given one, for example in Answer instructions, and never makes up a link or an address.

Let Yaplet try first

Let Yaplet try first is off unless you switch it on. Switched off, a visitor who asks for a person from the menu (Request agent) or with the Talk to support team button is connected to your team in one press, and one who tells the AI they want a person gets the Yes/No question described above.

With it on, a visitor who asks for a person from the menu or with the button is not connected straight away. If they haven't said what they need, a Request a person card asks first — "Tell us briefly what you need, so we can help faster." — and they can't skip it: pressing Request agent or the button again only puts the cursor in its box (the × still closes the card). Once they've written it, or when their request already names the issue, the chat shows Preparing your request… for a few seconds while Yaplet checks. Then exactly one of two things happens:

  • Yaplet can settle the issue — it answers right away in the chat, as a normal answer (typed out, with links and product cards), with no person buttons next to it.
  • It can't settle it — the visitor wants a status changed or something only your team can change (a stuck or late order, a held payout, a pending review, a refund), or your knowledge only says "contact support" — the visitor is handed to a person at once, and your team gets what Yaplet found as a private summary in the Inbox. If Yaplet's answer then finds nothing, it hands over at once too. A visitor who typed the request and for whom Yaplet judged a person is needed still gets the Yes/No question.

Asking for a person again after Yaplet's answer always reaches a person. A request right after one of Yaplet's answers (within 30 minutes, with nothing new written) goes straight to a person, with no card and no check: a typed request gets one short line saying someone from the team will join, and the ⋮ menu hands over without a line. A visitor the AI can't answer right now (your plan's AI allowance is used up with no bought credit, their AI is paused, or they reached the daily limit) is connected at once as well. If nobody from your team is online, "a person" means the usual offline notice and the leave-a-message card.

Every check is listed on Reports as a Hand-off check.

A check costs the same as an AI answer. Each one is billed like an AI answer and counts toward the visitor's daily AI answers (see Usage limits below), and so does the answer Yaplet then gives the visitor.

Usage limits

How many answers this AI agent gives one person in 24 hours is a brand setting, not one of the agent's own settings — the same numbers cap the phone agent too — so it lives on Brand → Brand settings → Usage limits (permission Brands & chat widgets). The Answers section links straight there. What the caps are and what happens when a visitor hits one is described on Brand settings.

Danger zone

Delete AI agent removes the agent, its conversation starters, its suggestions and its report history. This cannot be undone.

Your knowledge is not affected. Knowledge bases, documentation, files, website pages, Q&A entries and products all belong to the brand rather than to the agent, and stay exactly as they are. Add another AI agent later and it answers from the same material. There is no list of chatbots to go back to, so after deleting you land on the brand's Yaplet screen with the offer to add one.

Settings that live on the brand

Five settings describe the company and its content rather than this one agent, so they belong to the brand — and they apply to everything that answers for the brand: chat, phone calls, the inbox's suggested replies and AI social posts alike.

SettingWhere it is
Company name and WebsiteBrand → Brand settings → General
Usage limitsBrand → Brand settings → Usage limits
Knowledge languageSet on each source individually, on Brand → Knowledge
Search depthBrand → Brand settings → Advanced (Standard or Deep)
Privacy maskingBrand → Brand settings → Advanced
There is nothing to wire up. A brand owns one chat widget and one pool of knowledge, and Yaplet answers for that brand, so its knowledge always belongs to the right AI agent. The widget itself is at Brand → Chat widget.
Several languages work at once. Each knowledge base and documentation set carries its own language, set on Brand → Knowledge, and the brand's language is only the fallback for things with none. The agent searches in every language the brand's content is written in, so a brand with an English and a Hungarian knowledge base keeps full accuracy in both. This only affects how questions are matched to content — the agent always replies in the visitor's own language.

© 2026 Yaplet. All rights reserved.