Skip to content
E-commerceFind, track and return orders.BankingSpending, cards and transfers.TravelConnections and rebooking.TelecomBills and roaming passes.InsuranceClaims from one sentence.
AnswersTheir own data, shown as your app’s cards instead of a paragraph.ActionsTasks done through your API, confirmed with a tap or undone with one.VoiceThey say what they need and the app does it.MemoryRemembers what each person prefers and uses it the next time they ask.Your brandLooks like your app on iOS, Android and the web.Setup by your agentYour coding agent adds it and fills in the console.EngageMessages people asked for, written for you and sent with one tap.AnalyticsWhat people did, what they asked, and what got solved.Human supportYour team answers inside the app, with the conversation in hand.ImproveAnswers ready for what it missed, and requests counted by who asked.Why RendelA chatbot talks about your app.How it worksThe six steps of a turn.SecurityWhat nothing happens without.CompareNext to the alternatives.
DocsEverything, in eighty pages.FlutterThe package, on iOS and Android.WebOne script tag.React NativeSoonPlanned.SwiftSoonPlanned.KotlinSoonPlanned.
BlogWhat we are working out.GlossaryThe words, one sentence each.
Answers

“Where’s my order?” gets the order, not a help article.

Your customers ask in their own words and get back a page of your app’s own cards: their order, their return, your policy, built from their own account and your help content. Rendel is the Mobile and Web SDK that puts this assistant in your app, and your team runs it from one console.

Try the consoleBook a Demo

Halden’s console, an invented store with Rendel running. Yours for two hours, no sign-up.

Mira asks where her order is

Halden’s app reads her order through its own API and answers with the order itself.

list_ordersSucceeded

The problem

Your customer asked about their order. The help bot answered with an article.

Most AI help widgets read your help centre, not the customer’s account. So “Where’s my order?” gets a paragraph about delivery times, the customer goes looking for the right screen, and some of them write to your team instead.

It cannot see the person.

A widget that reads only articles gives everyone the same answer. The order, the size and the delivery window stay on another screen of your app.

It answers in a bubble.

A paragraph and a link in someone else’s chat window, where your app already has an order tracker it could have shown.

Two tools, and a gap between them.

One tool answers the FAQ, your support queue answers anything about an account, and nobody sees which questions fell between the two.

What 5,728 customers told Gartner

14%
of customer service and support issues were fully resolved in self-service.
45%
of customers who started in self-service said the company didn’t understand what they were trying to do.
43%
of cases, customers in self-service could not find content relevant to their issue.

Gartner press release, 19 August 2024.

How it works

From a question to a page, in five steps.

  1. They ask in their own words.

    Typed, or tapped from the suggestions on the opening screen. One sentence is enough.

  2. Rendel reads what the answer needs.

    Their account, through your app’s read actions, run as them with no confirmation needed, and your knowledge, searched by meaning.

  3. The answer is drawn as your app’s cards.

    Up to twelve blocks from the catalog, each shown once it is checked: the order tracker, the product cards, the policy in three points.

  4. The next step is one tap.

    A question to go on, a product’s details over the page, or your app’s own screen.

  5. Your team sees how it got there.

    The Test page and every conversation show the actions it ran and the knowledge it read for each answer.

Pages

Every answer is a page of its own, not a bubble in a thread.

Customers read the answer the way they read the rest of your app: the answer first, the detail under it, and the next step at the foot.

A fresh page for every question

The question is written over the screen, the old page dissolves and the new one builds block by block. The question stays as one faint line on top, so every page says what it answers.

It shows what it is doing, then writes as it goes

While it works, a pill in the microphone’s place says what it is doing. Text streams word by word, and every other card holds its place until it is checked, so nothing half-drawn reaches the screen.

  • Reduced motion respected

The answer first, then the reasons

A full answer opens with the answer in a sentence or two, the phrase that matters marked, then short titled sections and a verdict when it recommends something. Rendel is told to write this way, and a short question gets a short answer.

The next question is one tap away

Each answer ends on a few questions in the customer’s own words. When the model leaves them off, the server writes them when it can in time, except on a page that waits for the customer, like a form or a confirmation.

Back to an earlier page, from memory

Back and forward move through the pages this conversation opened, from memory, with no new request. Asking from an older page starts a new path from there.

Product details over the page, and the page still there

Tap a product or a list row and its details slide up: picture and name at once, then a summary, tags and up to two actions. Close it and the page is as it was; open it again and the details are already there.

It ends on your app’s own screen

When the right place is a screen your app already has, the answer names it and opens it, and Rendel shrinks into a bubble while it is open. It uses the screens your app registers.

Ask whether an answer helped

Turn on “Helpful?” and a thumb up and down sits under each answer, counted in Analytics. A thumbs-down marks the conversation as not resolved, so it reaches your team’s numbers.

  • Off by default
  • Mobile SDK 0.49+ · Web SDK 0.26+
Your cards

Drawn with your app’s own components, not written as text.

37 kinds of block, from product cards and receipts to maps and passes, drawn as native widgets in your mobile app and as page elements on your website, in your look.

37 kinds of card for what people ask about

Product cards, order trackers, receipts, comparisons, forms, maps, passes, charts and more, on every plan. Rendel draws 35 of them; the confirmation card and the Undo line are built only by the server.

  • Every plan
See every block

An old app version never shows a broken card

Each build of your app says which cards it can draw, and Rendel offers only those; a card it cannot draw arrives as its one-line sentence. A card that fails the server’s check is sent as that sentence too, never as a wrong card.

  • Checked by the server

Numbers people act on come from your own data

A number to call, an address, a map pin, a calendar time, a pass code or a value to copy must equal a field your app’s actions returned, or a contact channel you listed. A number planted in a help article is never one tap away.

  • Checked by the server

Your product photos, from your catalogue

When your data has a product’s picture, the card shows it. Rendel is told to copy picture links exactly as your API returns them, and only secure https links are drawn.

Your look, down to the card style

Cards with a shadow, an outline or flat; buttons filled or outlined; products as tiles, a list or photos; order tracking down, across or as a bar; your colours and fonts. Set in Appearance and published with the rest.

How Appearance works

Any card, drawn by your own widget

In a Flutter app, any one catalog card can be drawn by your own widget while the rest stay Rendel’s. New card types are not offered: you restyle the ones in the catalog.

  • Flutter apps
Their account

Their orders, returns and plan, read as them: no order number needed.

Rendel reads what the signed-in person can already see in your app, through your app’s own actions, and answers from that.

Their account, read as them

Rendel calls your app’s read actions, either inside the app with its own session or from Rendel’s server with the person’s own token, used for that call and dropped. Reads run at once, without asking them to confirm.

It knows who is asking, and what time it is for them

The signed-in id and the details your app sends, like a first name or a plan, reach the answer. Each question carries the person’s own clock, so “today” is their today.

  • Verified identity, optional

“My last order” and “this one” are enough

Nobody needs an order number: Rendel finds the record with a read action and asks which one by name when several match. When your app shares the screen it is on, “does this one run small?” means the product they are looking at.

It says what the data says

Rendel is told to name a status as your data names it, mark a step done only once reached, work totals out exactly and promise no refund your data does not show. When a size is sold out, it shows the sizes that are in stock.

Two questions in one message, both answered

Rendel is asked to answer every part of a message, in order, and to say which part it could not do.

A photo or a file says it faster

Customers can attach a photo, a PDF or a text file to a question, up to 20 MB, and the answer reads it. Files are not masked, and attachments are off while you mask names and addresses.

Your knowledge

Your policies and help pages, answered in your own words.

What a customer asks that is not in their account comes from what you give it: your website, documents and FAQs, searched by meaning on every question.

Try it in the demo console

Five ways in, from one paragraph to your whole website

Paste text, link a page, import a website of up to 50 pages, upload up to 20 files at a time, or add FAQ pairs. Linked pages are read again every week.

  • PDF, Word, HTML, Markdown, CSV and more

Found by meaning, not by matching words

Each source is split into passages and indexed by meaning within a few minutes, so “can I send it back if it doesn’t fit?” finds your returns page. Each question searches only this app’s knowledge, and a short follow-up is read with the message before it.

No made-up policies

Rendel is told to take prices, deadlines, hours and terms only from your knowledge or your app’s data. When nothing covers a question, it says it does not have that here and points to a contact channel you listed.

Your help content is fenced off from its rules

Your documents are fenced off and read as information, and Rendel is told never to act on them as instructions.

See which parts would answer, before anyone asks

“Try a question” runs the search a real chat runs and marks each part Used or Too weak. Rendel answers from the parts marked Used; the rest are too far from the question.

Keep every source right, and see who changed it

Take a source out of service at once on test and live, put it back, fetch a page again, edit it in place or delete it. Each source shows whether it is used, out of date or failed, with who did what and the answers that quoted it.

Knowledge that fills itself in

When you add your app, Rendel reads your website and store listing; the setup agent writes sources from your code; live offers become a “Current offers” source; and every answer typed in Questions to answer becomes a source.

Every question it could not answer, listed for you

Knowledge counts the questions your sources don’t answer, and Questions to answer lists each one from the last 30 days with how often it was asked and answers ready to tap. Each answer becomes a source.

How Improve works

It goes live once it has something to answer from

Two of the five Go live checks are about answers: at least one knowledge source Ready, and at least 80% of the check questions answered from your information or your app.

First words

The first question asked in one tap.

Set what the opening screen says, or start from lines Rendel writes from your own app’s actions and knowledge.

A welcome screen that gets the first question asked

A headline, a line under it and up to four suggested questions, set in Experience or in your code. Leave one blank and the SDK’s own words show, translated.

Ready-made lines from your own app

Experience offers a few ready rules, topics and welcome lines for each field, each naming the action or knowledge source it comes from. They are not charged as messages.

In place

Fix a detail, and the card changes where it is.

Customers correct one thing and keep the rest: a size on the card, a field in the form, an item in the bag.

Change a size right on the confirmation card

On a confirmation card, the customer can change a value you marked as changeable, like a size, and the server checks the whole request again before redrawing the card. Ids and amounts your app works out are never changeable.

How Actions work

A form that keeps up with the conversation

Say “it’s too small” and the return form opens with that reason chosen; say “actually, store credit” and the same form changes that one field and keeps the rest. It asks only for what is still missing.

Forms never ask for passwords or card numbers

A form asking for a password, a one-time code, a card number, a CVV, an IBAN or a national ID is never shown, whatever the instructions said. File fields appear only where you turned them on.

  • Checked by the server

A bag that updates in place

A cart, a list being narrowed or an order’s status is redrawn where it is instead of piling up, and choices already made stay chosen.

A tap means exactly what was tapped

Tapping a size, a time slot or a choice sends that option itself, and the server checks it against the card it drew, so the answer never guesses. An option that has gone or was already answered is turned down, in the customer’s language.

  • Checked by the server
Languages

Answered in the language each customer writes in.

Serve every market from one setup: answers follow the person, and your help pages need not be written twice.

In the language they write in

Each customer is answered in the language they write, switching when they switch. A tap or a “hi” that shows no language gets the phone’s language if it is on your list, otherwise the first one listed.

Help pages in one language, answers in theirs

Knowledge written in English answers a question asked in Turkish: each search also runs in English beside the person’s own words. Rendel is told to give titles and labels from your data in their language too.

Written in the wrong language? Written again

When your app names its languages and an answer comes out in the wrong one, it is held back and written again, once.

Buttons and labels in six languages

The SDK’s own words, the status pill, the confirm buttons and the composer, follow the answer in English, Turkish, German, French, Spanish and Arabic, right to left for Arabic on the web. Elsewhere they stay English.

Your voice

Answers in your brand’s voice, and changes you can take back.

One editor decides how it speaks, every change is a draft until you publish it, and Analytics shows what each version did.

Your brand’s voice, in one editor

Who it is, how it greets, how it talks and what it won’t do, in the order a customer meets it: a tone from four presets or your own words, three answer lengths, rules it always and never follows, and up to 20 topics it declines.

Draft, try, publish, and take a publish back

Saving writes a draft the Test page uses from its next turn; Publish puts it live from each customer’s next question. Every published version can be restored to the draft or sent straight back to live.

See whether an edit helped

Analytics breaks answers, the share answered with nothing of yours behind them and the ratings down by persona version, so you can tell whether v3 did better than v2.

Rules that hold, whatever the persona says

The persona shapes how it talks, not what it may do: only your registered actions run, changes wait for the customer’s tap on a card the server builds or run with an Undo where you allow it, and only catalog cards are drawn. It is told not to reveal how it is set up, to stay courteous and never to answer abuse in kind.

Your team sees every answer as the customer saw it

The Test page and each person’s conversations draw the same pages, beside the actions it ran and the knowledge it read. Test conversations are never billed.

Around the answer, each on a page of its own.

A way to a person

When someone needs your team, the answer shows only the contact channels you listed, and never says when someone will reply.

Read more: Human support

Your own feedback question

A rating card, at most once per conversation and only after the job is done.

Read more: Analytics

Remembers what they prefer

Forms and cards arrive filled in from what each person told it before.

Read more: Memory

From an answer to a task

Changes run through your API after a tap on a card the server builds or, for what you mark easy to take back, at once with an Undo.

Read more: Actions
In the demo

Ask Halden, then change what it knows.

Halden is an invented shoe store with Rendel inside. In the demo its answers are written in advance and played by the real SDK; the console, the knowledge and the versions around them are the real console, yours for two hours.

  1. Open the console.

    Try the console, then Start. Your copy opens on Test, with Halden’s questions ready to tap.

  2. Ask about a product.

    Tap “Find white sneakers under $130 in size 42”. The phone draws three pairs, how to choose and where to go next, and This turn shows search_products ran.

    Find white sneakers under $130 in size 42

    Three white pairs under $130 come in size 42. The Court Low is the one most people pick, with 4.7 stars from 1,284 reviews.

    Best sellerCourt Low$118★ 4.7
    Dune Slip-on$74★ 4.4
    Court Mid$126★ 4.6

    search_productsSucceeded

  3. Ask about an order.

    Ask “Where’s my order?”. list_orders runs, and the page is order H-20418’s tracker with a card that opens the order’s own screen.

  4. Open Knowledge.

    Your assistant › Knowledge holds Halden’s 13 sources. Try a question to see which would be used. In the demo copy this ranks by shared words on the same scale; your app’s knowledge is searched by meaning.

  5. Add what it does not know.

    Add source › FAQ: “Do you deliver on Saturdays?”, with your answer. It shows Ready.

  6. Ask it on Test.

    Back on Test, ask “Do you deliver on Saturdays?”. The answer is the one you wrote, and Knowledge it read names your new source.

    This answer

    • Do you deliver on Saturdays?
    • Answered

    Knowledge it read

    Delivery days

  7. See it as a shopper.

    Open Halden app from the demo bar: the same store in the real web SDK. Tap a product for its details, then close them and the page is still there.

Only the customer’s side? Halden’s store app is at Industries › E-commerce.

Try it in the demo console

Halden’s console with thirteen knowledge sources, three versions of its voice and the Test page, yours for two hours with no sign-up.

Try the console
Compared

Their account and your help content on one page, where a help widget keeps them apart.

A chatbot widget answers from your help centre in a window of its own. Rendel answers from the person’s account and your help content in the same page, inside your app.

WhatRendelA help-centre chatbot
What the customer gets backA page of your app’s own cards: the order tracker, product cards, a receipt, a formMessages and cards in the vendor’s messenger
Where the facts come fromThe signed-in person’s account, through your app’s read actions, and your knowledge, in one answerHelp-centre articles; account data through connectors or procedures set up in the vendor’s tool
A number to call, an address, a tracking codeMust equal what your API returned or a channel you listed, or it is sent as textDepends on how it is set up
When nothing covers the questionSays it does not have that here, points to a channel you listed, and lists the question for your teamDepends on the tool; often a handover or a ticket
How it looksYour colours, fonts and card styles, drawn natively in your appThe widget’s own style, themed within its options
Checking it before customers doThe Test page shows the actions and knowledge behind each answer, never billedA preview inside the vendor’s tool
Where it runsInside your Flutter app on iOS and Android, and on your websiteOften also email, voice and social, with native iOS and Android SDKs
What you pay forEach answer Rendel writes; what people type is freeOften each resolution or outcome, sometimes with helpdesk seats

Building it yourself on a generative UI library? Rendel vs building it yourself. Prices next to Fin and Lyro: Pricing compared.

Control

Who can change what, what is kept, and what never happens.

Your team decides what Rendel knows and how it speaks; the customer decides what to ask and what to attach.

Who can change what

WhatWho can
Add, edit, take out of service or delete knowledge; edit and publish Experience; Appearance
  • Owner
  • Admin
  • Developer
Read conversations and Questions to answer; run the Test page
  • Owner
  • Admin
  • Developer
  • Conversation reviewer
Turn on “Mask names and addresses from your actions”
  • Owner
  • Admin
Change retention, erase a user, export data
  • Owner
Answer people in Support, and nothing else: no knowledge, no Experience
  • Customer support

What never happens

  • A form never asks for a password, one-time code, card number, CVV, IBAN or national ID.
  • A number to call, an address, a map pin or a code to copy never comes from a help article or the model’s own words.
  • The person’s token for a console action is never stored, logged, shown in the console or put before the model.
  • An app build is never sent a card it cannot draw.

What is kept, and where

  • Conversations, messages and action records for your app’s window: 365 days unless an owner sets it between 30 and 730 days.
  • Knowledge until you delete the source. Uploaded files with their conversation; one never sent is deleted after a day.
  • Masked values deleted 24 hours after a conversation’s last message.
  • Stored in Frankfurt (AWS eu-central-1). Answers are written by Anthropic’s model, which does not train on them under its terms. Knowledge search uses OpenAI embeddings in the United States, which see masked questions and your knowledge text. Map tiles are loaded by the customer’s device from OpenFreeMap through Cloudflare.
  • Emails, phone numbers and card numbers that users type are masked before the model sees them.
  • Erase one person from Settings › Data & privacy › Erase one user.

What your team can switch off or narrow

  • Take any source out of service, at once on test and live.
  • Mask names and addresses from your actions; off by default, and file attachments are refused while it is on.
  • “Ask whether an answer helped”, off until you turn it on.
  • Files in forms, off until you turn them on.
  • What it must never do, and topics it won’t discuss.
  • Which languages it answers in when a message shows none, and which count as right.
  • Roll back live to any published version.

What the customer decides

  • What to ask, in their own words or with one tap.
  • Whether to attach a photo or a file.
  • Less motion: with reduced motion on, nothing reveals, dissolves or drifts.
  • A change waits for their tap on a card the server builds or, where you allow it for something easy to take back, runs with an Undo they can tap.

More: Security summary and What we store and for how long.

Turning it on

Answers are on from the first question. Account answers need your read actions.

Every plan has all of it, from the free Pilot with 400 messages a month.

  1. Add the SDK.

    The Flutter package rendel or the web script tag, with your app’s publishable key, or let your coding agent add it.

    AI setup

  2. Say who is signed in.

    identify the person and hand over their token, then register read actions in the app or connect endpoints under Capabilities › Actions.

    Sign-in and identityConnect an endpointThe signed-in user

  3. Register screens.

    Optional: so answers can end on them, and share app state so “this one” means what is on screen.

    Screens

  4. Add knowledge.

    At least one source; Go live needs one Ready.

    Knowledge

  5. Write the Experience.

    Name, languages, tone, rules, welcome; save the draft, try it in Test, publish.

    Persona

  6. Set the look.

    Card styles and “Ask whether an answer helped”, in Appearance.

    Theming

  7. Pass Go live.

    Run check questions (80% needed) and the five checks.

    Go live

Which SDK

Any current SDK draws everything on this page. Today that is Mobile SDK 0.82.0 and Web SDK 0.57.0.

For apps on older builds, each part arrived in:

PartMobileWeb
Answer pages and status pill0.36.00.13.0
Product details0.35.00.12.0
The question line0.46.40.23.5
“Helpful?”0.49.00.26.0
SDK words in six languages0.67.00.45.0

What teams ask before they start.

  • Yes, through the read actions you connect. They run inside your app with its own session, or from Rendel’s server with the person’s own token, used for that call and dropped, never stored or shown to the model. Reads run without asking them to confirm, and Rendel sees only what those actions return.

  • A page of its own, not a chat bubble: the answer in a sentence or two, then up to twelve cards from a catalog of 37, such as an order tracker, product cards or a receipt, and questions to go on at the foot, all in your app’s own look.

  • Not one a customer can tap. A number to call, an address, a map pin, a calendar time, a pass code or a value to copy must equal a field your actions returned, or a contact channel you listed; otherwise the card is sent as plain text. Rendel is also told to show only what your data holds.

  • When nothing in your knowledge or your app’s data covers a question, Rendel is told to say it does not have that here and point to a contact channel you listed. The question then appears in Questions to answer, where your answer becomes a source.

  • Pasted text, a linked page read again every week, a whole website of up to 50 pages, files up to 20 MB (PDF, Word, HTML, Markdown, CSV and more) and FAQ pairs. A scanned PDF needs OCR first. Each source is indexed by meaning within a few minutes.

  • The language each customer writes in, and it switches when they do. The languages you pick decide the language for messages that show none, like a tap, and which languages count as right. Its own buttons and labels come in English, Turkish, German, French, Spanish and Arabic, and stay English elsewhere.

  • No. Each build tells Rendel which cards it can draw and is offered only those. A card a build cannot draw, or one that fails the server’s check, arrives as its one-line sentence, so nobody meets a dead button.

  • Yes, within the catalog. Appearance sets colours, fonts, card and button styles, and product and tracker layouts, and a Flutter app can draw any one card with its own widget. New card types cannot be added.

  • On the Test page: it draws each answer as the customer would see it, beside the actions it ran and the knowledge it read, and it runs your draft. Knowledge › Try a question shows which passages would be used. Test conversations are never billed.

  • Only when the answer includes a help-article card, which names its source. Every answer records the sources behind it in its conversation, so your team sees them in the console.

  • Emails, phone numbers and card numbers that users type are masked before the model sees them. Conversations and knowledge are stored in Frankfurt (AWS eu-central-1), conversations for 365 days by default. Anthropic’s model writes the answers and does not train on them; OpenAI embeddings in the US search your knowledge.

  • By message: one answer Rendel writes is one message, and what people type is free. Every plan has all of it, from the free Pilot with 400 messages a month. An answer whose model cost passes $0.02 counts one more message for each $0.02 started. Test conversations are never billed.

Related

Put it in your app this week.

Start free on your own, or tell us what your customers do most and we’ll show you that exact workflow, working, in a tailored demo.

  • PilotFree400 messages a month
  • Starter$199 / month4,000 messages a month
  • Growth$799 / month20,000 messages a month
Start with the PilotBook a Demo
Every plan, and the estimatorOr try a live store’s console first, yours for two hours

Prefer email?

sales@rendel.ai

Prefer to write first?

Tell us about your app and we’ll show the right workflow in a tailored demo.

Help us tailor the demoOptional

We use this to get in touch about the demo and, if a partner introduced you, to credit them. We never sell it. See the privacy policy.