Add knowledge to your brand

Updated May 22, 2026

Where your knowledge lives now

You do not attach content to a chatbot any more. Everything the AI can read belongs to the brand, on one screen: Brand → Knowledge, headed What Yaplet answers from. Whatever is listed there is searched by the brand's chat AI, by its phone agents and by the reply suggestions your team sees in the inbox — in a single search, with nothing to connect up. AI Social Posting and Newsletter Autopilot work differently: they only pick what to write about from the sources ticked for them on the brand's Content sources page. Those ticks never change what Yaplet answers.

One screen, one group per kind

At the top, a summary shows your Public help site, how many Sources you have, how many searchable pieces Yaplet has stored from them (Indexed) and when your website was last read (Website last read). Press Add knowledge to open a window that asks what Yaplet should learn from. Under Teach Yaplet it offers Website, Files, Text snippet, Question & answer and API tool. Under Help content customers can read too it offers a Knowledge base and Documentation. A choice you cannot use stays on the list with the reason: "Ask an admin for access to add this.", or "Available on a higher plan." for Documentation on a plan that does not include it.

Below the summary, under the heading Sources, what you added sits in groups you can open and close, one per kind and with no tabs: Knowledge bases & documentation, Website, Files, Text snippets, Questions & answers and API tools. A group appears as soon as you have something of that kind. Knowledge bases and documentation sets are things your customers can also read, if you choose to put them on your public web address. Everything else — websites, files, text snippets, questions and answers, and API tools — is knowledge only: Yaplet answers from it in chat, on the phone and in your team's suggested replies, and your customers never see it. (Social posts and newsletter autopilots only pick topics from these sources when they are ticked on Content sources.)

Publishing indexes it straight away

When you publish a help article, it becomes searchable by the AI immediately. There is nothing to attach, no chatbot needed, and no widget needed. Edit an article and only that article is re-read; unpublish it and it is removed from what the AI can find.

Is it working? Status and pieces indexed

Every row tells you whether the AI stored anything from that source, and that is the fastest way to find out why a question came back empty. Rows for websites, files, text snippets and questions and answers carry a status badge:

  • Ready — stored and searchable.
  • Not indexed — nothing was stored from it, so Yaplet cannot answer from it yet.
  • Reading… — still being processed.
  • Failed — files, text snippets and questions and answers only: the last attempt to store it did not succeed. Hover for the reason.
  • Warning — read, but not cleanly. Hover for the reason.
  • Stopped — websites only: the site's list of pages could not be read for 14 days, so the hourly checks stopped. Yaplet keeps answering from what it learned; choose Check now to try again.

The line of facts under each name also says how many searchable pieces Yaplet stored from that source (for example 12 pieces indexed), or Not indexed when nothing was stored. One article usually becomes several pieces, and only published ones count. A knowledge base or documentation row shows the same count, or Not indexed yet, but no status badge. Once a group has more than ten rows, a search box appears above it (with a status filter too, except for knowledge bases and documentation), and a group that has a source needing attention opens by itself and says so in its header, for example 1 needs attention. If a website has stopped after repeated failed reads, a notice also tells you as the page opens.

Which knowledge base is on your public site

A brand can hold up to five knowledge bases and five documentation sets on any paid plan (one of each on the free plan), and the AI reads all of them. But only one knowledge base and one documentation set can sit on the brand's public web address at a time. Each row carries a badge, On public site or AI only, and the row's ⋯ menu offers Show at (your address) on a row that is not live, and Visit the live page and Take down from the public site on the one that is. In the summary at the top, the Public help site cell shows the address with a Live pill, or Nothing on it when the brand has an address but nothing is served there, or No address with a Set an address button when it has none yet.

Taking one down removes the public page and nothing else — the articles are untouched and the AI keeps answering from them. The first knowledge base you create in a brand goes live automatically; later ones start out as AI only. If the brand has no web address at all, the summary says so, and its Set an address button takes you to Brand settings.

One word of vocabulary, because it saves confusion: in Yaplet, publish means one thing only — the per-article switch that decides whether that single article exists. For a whole knowledge base the words are On public site, AI only, Show at… and Take down from the public site.

What language a source is written in

Each knowledge base and documentation set has its own Language, set inline on its row. Point at it and the screen explains it: "The language this is written in. It decides how Yaplet searches the text — it does not translate anything." A website chooses its languages in Manage pages. The brand's own Language, in Brand settings → General, does two jobs: it is the fallback for things that have no language of their own — files, pasted text and question-and-answer entries — and it always counts as one of the three languages Yaplet searches at full accuracy, whether or not anything depends on it.

Copy or move something to another brand

If you run more than one brand, the row menu has two more options. Copy to another brand… duplicates the whole thing — every category, article or page, with its nesting, language and ordering — and everything arrives unpublished, so nothing goes onto the other brand's public site, nothing is indexed and nothing is billed until you publish it there. Move to another brand… hands the original over with everything it has already learned, costs nothing, re-indexes nothing, and arrives as AI only. If you are moving the one currently on this brand's address, the dialog warns you in red that the address will have nothing behind it afterwards.

Deleting a source

Deleting a knowledge base or documentation set is a single action with a counted confirmation — it tells you exactly what goes ("84 articles and their categories") and that everything indexed from it goes too. If you are deleting the source currently on your public address, it says that as well. There is no undo. A website, file, text snippet, question and answer or API tool asks the same in one line, for example "Delete “Return policy”? Everything Yaplet learned from it is removed at the same time."

Add your website — let Yaplet read it

Press Add knowledge → Website, type your Website address and press Add. Your home page teaches the whole site, a sitemap address teaches that sitemap's list of pages, and a deeper address teaches only that page. Each site is one row. Yaplet reads it in the background — a few minutes, usually — and the row shows how far it is ("12 of 40 pages learned") and its status: Ready, Not indexed, Reading…, Warning or Stopped. If the site is already added, Yaplet tells you so instead of making a duplicate. A workspace can hold up to 5 websites — more with a paid plan or bought credit — and a website up to 2,000 pages.

To choose what Yaplet learns from a site, press Manage pages on its row. The pages are listed by section (Main pages, Articles and Other pages), and every switch applies at once, with no Save. A section you switch off there is off in every language. If the site comes in more than one language, you see one language at a time: the Language selector opens on the language we recommend — your brand's language if the site has it, otherwise the site's main language — and the other languages start switched off. Pages whose address has no language folder show up as their own entry in the selector, Pages with no language folder. Yaplet still answers visitors in their own language. To teach another language too, turn on Learn language pages for it; turning it off removes everything Yaplet learned in that language. Each page shows its state — Learned, Off, Waiting, No room or Failed, with the reason — See what was stored shows what was kept, and Open the page opens it on your site.

A new website is learned in a sensible order: your home page and the pages your menus and footer link to come first, then your product pages, then everything else. A site of 100 pages or fewer is learned whole. On a bigger site the blog and article sections start switched off in Manage pages, and you can switch them on whenever you like. On a website you added yourself, the row says "Check what I'll learn from this site" until you open Manage pages once.

A page that shows a product with a price becomes a product, with its price and stock, instead of page text, unless the website's Pages with product data switch says pages. You find the switch in Manage pages: on, "I learn them as products, with their prices."; off, "I learn them as pages, like the rest of your site." A new website starts with it on, and a website learned earlier stays on pages until you change it. A change applies as each page is read again. The products appear under Brand → Products; see connecting your products. Switching a product page off in Manage pages also removes its product.

If you pay for Yaplet from credit (your free 14 days' credits or credit you bought, with no plan), Yaplet shows what learning costs. In Manage pages, each section says how many pages are waiting and about how many credits learning them takes, for example "12 pages waiting · about 30 credits", and the website's row says "Keeping this site up to date: about N credits a week". When something you switch on would use 100 credits or more, or a quarter of the credits you have left, Yaplet asks first: Should I learn these pages?, with Learn it, Add credit and Cancel. Cancel puts the switch back.

Keeping a website up to date

You do not have to add a site again when it changes. Yaplet checks it every hour (the summary says Checked every hour): new pages are read, pages with a new date are read again (at most once a day), and a page that has been missing for two checks is forgotten. Every page it has learned is read again at least once every 7 days, and pages the AI uses in its answers are read sooner. Text that has not changed costs nothing. New pages are learned by themselves when their section and language are switched on; otherwise they wait until you switch on Learn new pages automatically on the site's row. Yaplet tells the account owner in its chat, at most once a day, what it learned.

To look at a site right now, choose Check now in the row's menu. A site can be checked once every five minutes, and Website actions (the ⋯ button in the Website group's header) has Read all again, which checks every site of the brand in the background. Pause automatic checks in the same menu stops the hourly checks, and Check now still works. If a site cannot be read at all, Yaplet keeps answering from what it already learned, emails the account owner after 24 failed hourly checks, and stops checking after 14 days.

Add files, text and questions & answers

Choose them in the Add knowledge window. Files asks you to drag and drop or choose PDF, DOCX or TXT files and to give the upload a Name (how it is listed on this page); press Upload. A Text snippet asks for a Title and the Text itself — a policy, opening hours, product details; press Add snippet. A Question & answer asks for a Question and the exact Answer Yaplet should give; press Add answer. A title, a file name or a question can only be added once. To change a text snippet or a question and answer later, open the row's ⋯ menu and choose Edit; Yaplet reads it again as soon as you save. The same menu has See what was stored, which shows exactly which pieces were kept, and Delete.

API tools for live data

Some answers cannot come from stored text — an order status, current stock, a booking. For those, an API tool lets the AI call one of your own endpoints while it is writing the answer. Add one with Add knowledge → API tool. It appears in the API tools group, where each row shows the tool's name, its method (GET or POST) and the start of its description, and a switch pauses or resumes the tool. The tool's window has four tabs — General, User inputs, API connection and Settings. On General, the Description matters most, and the wording counts: the AI reads that text to decide when to call the tool, so describe the situation, not the endpoint. An accordion called Tips for writing a good description helps. A separate, optional Response instructions box tells the AI how to present what your endpoint returns — which fields to quote, what a status code means, when to hand the conversation to a person — and it only reaches the AI once the endpoint has actually answered, so it costs nothing on turns where the tool is not used. The API endpoint, on the API connection tab, must start with https:// — API tools only call secure addresses, and any other address is refused when you save.

The other tabs, briefly. On General, Response format chooses between Text (Yaplet reads the response and answers in its own words) and Product cards. User inputs defines fields a visitor fills out in a form before the call (Add input field), and When to show the form decides whether the form is skipped when Yaplet has already filled in the answers. API connection holds the API endpoint, the HTTP method, the Secret key (every call sends its SHA-256 signature, so your server can check the call came from Yaplet), custom headers and the request body; the Test endpoint button at the bottom tries the call. On Settings, Timeout (ms), Retries (for calls that only read data) and Rate limit per minute control how the call behaves, and On failure decides what the visitor gets when it fails: Show custom message to user, Request human agent with message or Try to answer with other context.

The allowance is per brand: one API tool on the free plan, twenty on paid plans. Names may use letters, numbers, underscores and hyphens, up to 64 characters, and a name that only differs in capitalization is rejected as a duplicate.

On the tool's Settings tab, three settings keep it safe:

  • This call — Reads data only looks something up, so a call that failed may be retried. Changes data places an order, cancels or books something, and is never repeated, not even after a failure.
  • Calls per visitor per hour — how many times one visitor can make the AI call this tool in an hour. 0 = no limit.
  • Only for identified visitors — off by default. Switch it on for a tool that reads or changes one customer's own account, such as their orders or subscription. The AI then calls it only for visitors your website signed in through Yaplet.identify, in your website chat (see Identify your logged-in users with custom data); Facebook, Instagram, email and phone visitors never count. Only the user ID is verified, so let your server look the user up by {{ external_id }}. Everyone else is told in their own language how to sign in — the text you set in How visitors get identified on Brand → AI agent → Answers.

Every call to your server carries an X-Yaplet-Call-Id header: a unique id per call that stays the same when the call is retried, so your server can ignore a duplicate.

Who can teach Yaplet

One question decides it: can something pay for it? Adding or refreshing website pages, files, text, questions and answers and product feeds runs only when your free 14 days, credit you bought or a paid plan can pay for it. If none of them can, nothing new is learned or refreshed, nothing Yaplet already learned is removed, and the payment window opens so you can add credit or choose a plan. A help article you publish still publishes, and Yaplet learns it within the hour after you pay. On a workspace without a paid plan, Yaplet also keeps the last 500 credits for answers: once fewer than 500 credits are left, it stops learning new pages and products until you add credit (see Buy Yaplet credits for AI, voice, and email usage).

There is no fixed count of 250 items for everyone any more. A paid plan has no limit on pieces (a website can hold up to 2,000 pages). During the free 14 days, or with bought credit and no plan, there is a limit too: if you reach it, the website row says "You've reached your knowledge limit" and offers Write to us. Only older free workspaces that still have free monthly answers keep a plain 250 across the whole workspace. They are the only ones who see a count like "212 of 250 on the free plan" and the notice "Your free plan's knowledge is full", and one article is usually several pieces, so 250 is not 250 articles. When it is full, publishing new articles and adding new sources is paused until you remove something or upgrade.

Who can open this screen

The page opens for anyone holding any one of three permissions: Content sources, Knowledge base or Documentation. Creating or changing a knowledge base needs Knowledge base; a documentation set needs Documentation; and the groups for websites, files, text snippets, questions and answers and API tools are shown only to people who hold Content sources.

Product feeds are on their own screen

Product catalogs are deliberately not listed here — they live only at Brand → Products, so the same thing cannot be deleted from two places. They do not depend on the AI agent: a brand can hold products before it has a Yaplet. You choose how often each source is re-read, anywhere from every day to every 30 days, and your shop can also push changes to us the moment they happen. See connecting your products for how to set one up.

Next step

Check that what you added actually reached the AI: Test your chatbots before going live. Once real conversations have happened, see what is still missing: Find knowledge gaps Yaplet couldn't answer.

Did this article answer your question?