All work03 / 04

Self-initiated / Mobile and desktop / 2026

An onboarding agent that banks put inside their own apps. Their clients just talk to it.

The same moment in Kith’s default look, in Vessa and in Arkwell.
Type
Self-initiated project
Sector
Banking and wealth platforms, onboarding
Platforms
Mobile and desktop
Services
  1. Strategy and research
  2. Product design
  3. Brand identity
  4. Design systems
  5. Development

The person and the agent

The person is a busy principal: the least informed about most of her own file, and the only one who can answer a few parts of it. The agent reads, drafts, splits the questions by who knows what, contacts her people and assembles the file. She does three things herself: confirms who she is, answers the one declaration no one else may make, and holds the seal. Behind both is a third party, the institution, which dressed Kith in its own colours and receives the result.

A self-initiated project: no client, not launched, no users or results. Every person, company and figure in the screens is invented.

01 / Strategy and research

Ask the person who knows. Let the bank keep its own face.

The problem

Institutions must understand who a client is and who owns and controls the structures behind them. For a wealthy client that turns into a long questionnaire addressed to one person and forwarded by email to the people who actually know. Neobanks and investment platforms moving up-market have it worse: their onboarding was built for a mass-market sign-up form.

Two observations

The principal is rarely the person who holds the answers, and she already trusts her bank, not her bank’s vendors. So Kith is white label: the institution buys it and puts it inside its own app; the client meets an agent in a look she already knows.

Two audiences

For the institution: a finished, sourced file, a small console to receive it, and three dials to make Kith look like theirs: colour, type and corners. For the client: a conversation that asks almost nothing, shows what it already found and where from, and makes pointing at a source or a person a single gesture.

What chat costs

A form shows the whole task at once and lets you jump around. We kept that where it matters: a count (“38 of 41”) and the review carry scope; on desktop a side panel shows the whole file; every fact is a structured object with its source. And no chatbot habits: no greetings, no cheering, never two questions in a row.

Trade-offs

One institution per file, not a reusable identity passport. Drafts that people confirm, not answers the agent asserts. Provenance, not confidence scores. The agent called Kith in every host, so its manners stay recognisable. No compliance claims and no risk scoring in front of the client.

The white-label model: the institution, the client and her people, and where Kith sits between them.

02 / Product design

One conversation, two devices, three people, one institution.

Inside an invented neobank, Vessa, Elena is asked one thing: who already knows you? Every screen answers what to know, what to do and what makes it easier. The thread is short and bottom-anchored; what matters lives in the objects.

The UI rule, on one screen: what Kith found

  • Know

    That covers 23 of 41, and each fact shows where it came from.

  • Do

    Point Kith at the people who handle the rest.

  • Eases

    A square for every source, a circle for every person, and one tap to point.

Eight phone screens of one conversation inside Vessa, from "Who already knows you?" to holding the seal.

Mobile / 8 screens

  1. Kith inside Vessa asking the client who already knows her, with four ways to point.

    01 / 08

    Who already knows you?

    One question, inside Vessa’s app, in Vessa’s look.

  2. The client pasting a profile link and a company website into the conversation.

    02 / 08

    Pointing

    She pastes her profile and her company’s website.

  3. Two sources read, and three facts each marked with the source they came from.

    03 / 08

    What I found

    Kith reads both and shows the facts that matter, each with where it came from.

  4. The client naming four people as Kith splits the remaining questions between them.

    04 / 08

    Her people

    The rest sit with people. She names four; Kith splits the questions and tells her each sees only their own.

  5. The count at 38 of 41, with the four people and what each answered.

    05 / 08

    People answered

    Two days later: 38 of 41. Three are only hers.

  6. A passport scanned by an assistant, waiting for the client to confirm it with her face.

    06 / 08

    Is this you?

    She confirms a passport her assistant scanned, with her face.

  7. A declaration marked "Only you", with two answers to choose from.

    07 / 08

    Only you

    Does anyone in her family hold public office? Asked of no one else.

  8. The review of what Vessa receives, with the seal being held.

    08 / 08

    Hold to seal

    She reads what Vessa will receive and holds the seal.

Desktop / 6 screens

  1. The conversation on a laptop with the whole file in a side panel.

    01 / 06

    On her laptop

    The same conversation continues on her laptop; the face check goes to her phone.

  2. The family-office manager’s page with nine questions, six drafted.

    02 / 06

    A delegate is asked

    Her family-office manager opens a page in Vessa’s look: nine questions, six already drafted from the annual report.

  3. The manager adding a document as a source and handing one question to the accountant.

    03 / 06

    Points and hands on

    She drops in the 2024 accounts for the source-of-wealth answer and passes bank relationships to the accountant.

  4. The Kith console at the institution with the sealed file arriving.

    04 / 06

    The file arrives

    At Vessa, onboarding sees the sealed file arrive in the Kith console.

  5. An ownership answer opened to the source sentence and the person who confirmed it.

    05 / 06

    Provenance

    The ownership answer opens to the sentence on page 41 it came from, and who confirmed it.

  6. The console’s look settings: colour, type and corners for Kith inside Vessa.

    06 / 06

    The look

    Separately, Vessa sets how Kith looks inside its app: colour, type and corners.

Two phones: the client pasting a profile and a website as Kith reads them, and naming four people as Kith splits the questions.
The Kith console with the ownership answer opened to the sentence it came from and who confirmed it.

03 / Brand identity

Its own brand, a voice in someone else’s app, and a quiet endorsement.

The name is old English. Kith, from cȳþþ, “knowledge, acquaintance”, means the people who know you. A bank must know its client; the client already has people who do.

Kith’s own look is paper and ink with one seal: linen and umber with an oxblood accent, Newsreader for questions and the agent’s italic voice, Source Sans 3 for names and figures, and a symbol of a principal in seal with six people around her, open at top and bottom.

Inside a host, the host’s colour, type and corners take over and Kith keeps its manners: the voice (“Reading both.”, never “Great! I’m on it!”), squares for sources and circles for people, “Only you”, hold to seal, and one line beneath the conversation, “by kith”, in Kith’s own lettering at the host’s quietest colour.

Ground
Linen #F4EEE4
Ink
Umber #2B2520
Seal
Oxblood #7C2D24
Questions and voice
Newsreader
Names and figures
Source Sans 3
Four brand frames: cover lockup, the white-label model, the by-kith endorsement, and Kith inside two host apps.
The Kith symbol: a principal in seal with six people around her, open at top and bottom.
The by-kith endorsement and its rules.
Kith inside two host apps, in their colours, with its manners unchanged.
Voice: what Kith says and what it never says.

04 / Design systems

Theming is the system.

One variable collection with three modes, Kith (the default), Vessa and Arkwell, holds thirteen colour roles, three font roles and four radii. Text styles bind their family to those roles, so the same screen re-dresses itself when the mode changes. The agent’s name is one separate variable that every header reads.

The conversation is built from components with variants, used as instances everywhere. One rule still does a lot of work: circles are people, squares are sources, in every host.

Components

  • Header
  • Agent line
  • Client line
  • Point chip
  • Source
  • Fact
  • Person
  • Count
  • Divider
  • Identity
  • Only you
  • Reply
  • Review
  • Seal
  • Composer
  • Button
  • Mark
  • by kith

Rules

  • A host changes how Kith looks, not how it behaves.
  • Circles are people; squares are sources.
  • Provenance beside every answer, never a confidence score.
  • Hold to seal: same geometry and 1.6 second timing in every host.

From Kith’s tokens.json, rendered live in its own type

Colour roles, one column per theme mode

RoleKithVessaArkwell
groundPage and thread ground#F4EEE4#F6F6F3#EEF0F2
surfaceIn-thread objects, cards, composer field, panels#FCF9F4#FFFFFF#FFFFFF
inkPrimary text, agent voice, ink dots of the mark#2B2520#121417#0F1D2E
mutedSecondary text, labels, the by kith endorsement#7A7065#686C73#566273
faintPlaceholder, neutral status dot, connector lines#B9AE9F#A9ADB3#9AA4B1
lineCard strokes, row dividers, seal track#E3DACC#E7E7E2#D8DDE3
accentThe one primary action, seal wax and arc, centre dot of the mark, source letters#7C2D24#0E6B5C#0F1D2E
accent-soft'Only you' tag, source marker ground, point button#EEDFD8#DCEFEA#E1E6EE
on-accentText and marks on accent#FCF9F4#FFFFFF#FFFFFF
positiveRead / answered / confirmed text and dot; person initials#56664F#2F6B4F#2E6A57
positive-softAnswered pill, person provenance marker#E2E6DA#E1EEE5#DDEBE5
selfThe client's message bubble#E9E0D2#121417#DCE3EC
on-selfText in the client's bubble#2B2520#FFFFFF#0F1D2E

Type

  • display-xlNewsreader 300 / 120px / 1 / -2%

    Who already knows you?

  • displayNewsreader 400 / 56px / 1.06 / -1.5%

    Who already knows you?

  • titleNewsreader 400 / 34px / 1.12 / -1%

    Who already knows you?

  • headingNewsreader 400 / 24px / 1.22 / -0.5%

    Who already knows you?

  • voiceNewsreader 400 italic / 18px / 1.45

    Who already knows you?

  • quoteNewsreader 400 italic / 20px / 1.4

    Who already knows you?

  • bodySource Sans 3 400 / 17px / 1.45

    Who already knows you?

  • messageSource Sans 3 400 / 16px / 1.4

    Who already knows you?

  • rowSource Sans 3 500 / 16px / 1.3

    Who already knows you?

  • agent-nameSource Sans 3 600 / 15px / 1.2

    Who already knows you?

  • body-sSource Sans 3 400 / 15px / 1.4

    Who already knows you?

  • labelSource Sans 3 500 / 13px / 1.3 / 2%

    Who already knows you?

  • captionSource Sans 3 400 / 13px / 1.35

    Who already knows you?

  • smallSource Sans 3 400 / 12px / 1.3

    Who already knows you?

  • buttonSource Sans 3 600 / 16px / 1.2

    Who already knows you?

Spacing

  • 4
  • 8
  • 12
  • 16
  • 24
  • 32
  • 48
  • 64
  • 96
Theme variables in three modes beside the chat components they drive.

05 / Development

A slice of the conversation, built live, in three looks.

Pointing at a source, the provenance of what comes back, and the held seal, in one scripted piece. Switch the theme to see the same agent dressed as Vessa or Arkwell: the conversation keeps its place and its manners.

Live / Point Kith at the website, then hold the seal. Switch the look at any time

Inside Vessa’s app: its colour, type and corners. Kith keeps its manners.

Kith

for Vessa Private

How it is built

Stack
One React component and CSS; the script runs on timers in the browser, no parsing, uploads or network
Theming
The three modes from Kith’s tokens.json applied as CSS custom properties on the phone; colours cross-fade, fonts and corners switch
Type
Newsreader, Source Sans 3, Plus Jakarta Sans, Libre Caslon Text and Work Sans, served from this site
Keyboard
Arrow keys move through the themes; Enter or Space on a chip; hold Space or Enter on the seal, Escape cancels; new lines are announced politely
Motion
Reduced motion: opacity only, and a countdown instead of the arc

Next project / 04 of 04

BRANDVERSE

A brand visual language builder that writes down the few rules that make a brand look like itself, makes AI-generated artifacts inside them, and points at the one thing that is off.

Read the case study
BRANDVERSE wordmark beside a laptop showing six brand rules and a phone showing a draft post with one thing marked off.

Talk to us

We love to talk design.

Tell us about your product, the people who use it and where it hurts. We are happy to talk design, user experience and how design can set you up for success.

We use these details only to reply to you. Nothing is shared or added to a list.