What Yaplet is
Yaplet is your AI agent. It answers your visitors in natural language, using whatever your brand knows — knowledge bases, documentation, uploaded files, pages of your website we read for you, question-and-answer entries and your product catalog. Each brand has exactly one Yaplet, at Brand → AI agent.
The important part to understand before you start: the knowledge belongs to the brand, not to Yaplet. The same material answers phone calls, writes the reply suggestions your team sees in the inbox, and feeds AI social posting. You add content once and everything reads from it.
The fastest way to start
AI setup runs for you right after you sign up. Give it your website address and it fills the brand in for you: company name, description, website, language, icon and logo; the chat widget's design and a few quick buttons about your business; your website added as knowledge (the first pages are learned within minutes so Yaplet can write test questions from them, and the rest follows in the background); and a few conversation starters, but only if your agent has none yet. It never touches your web address or your widget's name. You can run it again later from Brand settings → General with the Fill from website button, which fills in the brand's details and can read your website into the AI's knowledge. Let AI set up your brand from your website walks through the whole wizard, including the buttons that run parts of it.
Run it again later and you get a Keep or Replace choice for each part, so a description you wrote by hand survives a re-coloring. On a brand-new brand there is nothing to keep, so no choices appear. AI setup is paid like any other AI work: first from the credits of your free 14 days, then from credit you bought or your plan. If none of them can pay for it, a message says "AI setup needs credit or a plan." A preview that saves nothing still costs an AI call.
Step 1 — Create the agent
Open Brand → AI agent. If the brand has no agent, the page shows one message — "Nothing answers with this knowledge yet" — and one Add an AI agent button. Press it and the agent is created already switched on, answering from whatever the brand knows at that moment.
A brand can have one Yaplet and no more, and creating one uses an AI-chatbot allowance from your plan.
Step 2 — Give the brand something to answer from
All content lives on one screen: Brand → Knowledge. Press Add knowledge and pick what Yaplet should learn from: your Website, Files, a Text snippet, a Question & answer, an API tool, or a Knowledge base or Documentation for your public help site. What you add sits on the same screen, in groups you can open and close. Nothing is attached to the agent — you add it to the brand and the agent reads it.
Here is the reassuring part: publishing an article indexes it straight away. It does not matter whether an agent exists or whether anything is connected to a widget. You can fill a knowledge base today and add Yaplet next week without re-importing a thing.
Product catalogs are the one exception — they have their own screen at Brand → Products, and they no longer depend on the AI agent: a brand can hold products before it has a Yaplet at all.
Step 3 — Set how it behaves
Back on Brand → AI agent, the side menu has a section for each part of the setup. These are the ones to look at now:
- General — the agent's Name ("Shown to visitors in the chat widget and to you in the inbox."; it reads Yaplet and is locked, because renaming Yaplet is available on Enterprise plans), the Enable Yaplet switch, the Answers on switches (Website chat, Facebook Messenger, Instagram and Email — where Yaplet answers by itself; on a channel you switch off, messages go straight to your team and workflows keep running), a Phone row that says how many phone agents the brand has and how many are live (it opens the Phone section), and a read-only AI models note naming the models Yaplet uses (Answers and Quick tasks).
- Answers — three groups. Writing style holds the Tone (Friendly, or Formal for a voice that never uses emojis) and the Answer instructions ("Sent with every answer: rules, tone details, or facts Yaplet should always know."); the same instructions show up as Instructions in the Playground. AI model holds the Answer model, which you can pick when your plan lets you choose; it is the same setting as Settings → AI Models, so it changes the model for all AI in your workspace. How visitors get identified is the message that visitors who are not identified read when a tool needs them signed in.
- Human hand-off — every way a visitor can reach a person, what happens when Yaplet has no answer, and whether Yaplet tries to solve the issue first.
- Starters — the buttons visitors can tap when a chat opens. Add starter asks for the Button text, what happens When tapped (Ask Yaplet or Start a workflow), the Question for Yaplet and Show to visitors; a starter's menu has Move up and Move down to set the order.
- Suggestions and Reports — what Yaplet could not answer and how well it did; see Find knowledge gaps Yaplet couldn't answer.
- Danger zone — Delete AI agent. It removes Yaplet with its starters, suggestions and report history; your knowledge stays.
Changes on General, Answers and Human hand-off are saved with Save changes in the title bar (Discard puts back the last saved values), and leaving the page with unsaved edits asks first.
Three settings people look for here live elsewhere. Your company name, website and language are brand settings, at Brand settings → General — and Fill from website can write them for you from your website address. Search depth and Privacy masking are on Brand settings → Advanced. Search depth is Standard by default; Deep also searches with two more wordings of the visitor's question, so it finds content that is phrased differently, but answers come slightly slower, at a slightly higher AI cost. Privacy masking hides email addresses, phone numbers and card numbers in what visitors write before that text is sent to the AI provider; it is best-effort rather than a guarantee, so if the masking service is slow or unavailable the message goes out as written.
Step 4 — Decide what happens when it cannot answer
In the Human hand-off section, When Yaplet has no answer gives you two choices:
- Politely decline — the agent says it does not have information about that topic. It does not guess.
- Connect to a person — if one of your people is online, the visitor is connected to them right away, with no question asked first. If nobody is available, it still hands over: Yaplet tells the visitor that nobody from the team is available right now, and the website chat shows the card for leaving a message for the team.
The second option is grayed out until Visitor asks Yaplet for a person is switched on, in the same section. That section also holds the switches for “Request agent” in the chat menu and the “Talk to support team” button, which by default appears after six messages within the last five minutes. Let Yaplet try first is off unless you switch it on: with it on, when a visitor asks for a person, the AI first makes sure it knows what they need and checks whether it can solve the issue itself — it answers when it can, and connects them to a person right away when it can't. The line a visitor reads before being connected is Yaplet's own, said in their language, so there is nothing to write for it.
Step 5 — Try it
Open Brand → Playground, the first tab of the brand menu. On the left you see whether Yaplet is trained and, after AI setup, a few Test questions written from your website — each one shows the answer your site gives, so you can compare. Below are the Model (the same workspace-wide setting as Settings → AI Models), your Knowledge, the Actions Yaplet can take (workflows and API tools) and Yaplet's Instructions, with a Start from a template menu and an Undo changes button. On the right is a test chat that already answers with instructions you have not saved yet, so you can change a line, ask again, and save only when you like the result. Test chats stay out of your inbox and your reports. While Yaplet is still learning your website the chat says so (Still learning your website — answers get better in a moment.), and if website chat answers are switched off for this brand it tells you that visitors will not get these answers, though the test works anyway. Ask five questions you know the answers to. When you are happy, the Install on website button in the title bar puts the chat on your site.
To see the genuine widget the way a visitor does, use the live preview at Brand → Chat widget. That one is the real widget, so whatever you write there becomes a real conversation in your inbox.
Pre-launch checklist
-
The agent exists and Enable Yaplet is on
-
Brand → Knowledge lists at least one source that reads Ready or shows pieces indexed, and none that says Not indexed
-
Answer instructions written, or intentionally left blank
-
"When Yaplet has no answer" set the way you want it
-
Human hand-off settings reviewed
-
Tried in the Playground with at least five representative questions
Next steps
Add more content: Add knowledge to your brand. Tune the voice: Write a system prompt and tone for Yaplet. Control escalation: Control when Yaplet hands off to a human. After real conversations happen, see what it could not answer: Find knowledge gaps Yaplet couldn't answer.