How it works

From messages to meaning.

Every relationship has a shape that is hard to see from inside it. MOSAIC steps back and looks at the whole conversation at once, so the patterns you only half notice come into focus.

  1. 01 · Your messages

    Bring your chats

    • iMessageandWhatsApp
    • You choose which chats, nothing else is read
    • Private info stays anonymous
  2. 02 · The analysis

    MOSAIC analyzes them

    • Every message, both sides
    • Spots patterns and subtle nuances
    • Months or years at once
  3. 03 · The result

    Get your report

    • What keeps happening
    • The messages that show it
    • How sure each finding is

Then choose a report

Two reports, two questions.

Explore the reports →

Core

Reads a fixed, limited window of messages. A quick first look, and a good way to try a lens before going deeper.

In-depth

Reads every message, from the first to the latest. More history means clearer patterns, so findings are more accurate and more comprehensive.

How much to give it

More history, clearer patterns.

A pattern only counts once it repeats. MOSAIC counts messages per person, so both of you need to be writing.

Too little

<50

messages each

Most findings read “Not enough evidence”. There is not enough to tell a pattern from a moment.

Minimum

200+

messages · 1,000+ words each

Enough for a first read, and the least a relationship report needs. Expect more Moderate and Low confidence.

Good

1,000+

messages over a few months

Patterns have room to repeat through good weeks and bad ones, so more findings reach High.

Best

1 yr+

of history

In-depth reads all of it. Slow changes and rare patterns only show up here.

Best results when

  • It is a one-to-one conversation with the person the report is about.
  • Both of you write in it, not just one of you.
  • It covers ordinary days and hard ones: plans, disagreements, making up.
  • Most of your contact happens there, rather than in person or on calls.
  • It runs up to the present, so the report reflects where things are now.

Worth knowing

  • Photos, stickers and voice notes in a chat carry no text to read. Add a recording if you mostly talk out loud.
  • A tense month can look like the whole story. More history evens it out.
  • Very long histories: the report keeps your most recent messages up to what your plan includes.

Why not a chatbot

Where the technology differs.

Paste a chat into ChatGPT, Claude or Gemini and one general model reads it all at once and answers in one go. Fine for a recipe. Not for years of a relationship. Here are ten things MOSAIC does that a chatbot can’t.

The needle in the haystack.

The newest models can take in a whole history at once, but taking it in isn’t the same as using it. Hide one important fact in a long prompt and models regularly miss it, more often as the prompt grows and well before the window is full. Researchers call it the needle-in-a-haystack problem. Chat apps often search a long upload for snippets instead of reading all of it, and you’re rarely told what was skipped. That missed message might be the one that explains everything after it. MOSAIC reads every message, in stretches of conversation short enough to read closely.

  • MOSAIC reads every message in sections small enough to read closely. A chatbot can take it all in and still use only part of it.
  • MOSAIC reads each segment start to finish. A chatbot app often searches a long upload for snippets instead.
  • MOSAIC finds the one message that explains the rest. A chatbot can skip right past it, and won’t say it did.

SourcesModarressi et al., 2025·Chroma, 2025·Song et al., 2025

Pasted into a chatbot

One long prompt

Reading one long prompt…Pieces went missing, and the needle with them.

MOSAIC

Every message, in order.

48,210 of 48,210 messages readFound: the message from month seven.

Illustration. Based on long-context research on language models (Liu et al., 2023; Hsieh et al., 2024) and needle-in-a-haystack tests.

Built on research, not made-up labels.

Ask a chatbot what’s going on and it reaches for whatever words come to mind: distant, mixed signals, clearly cares. They sound right, but they aren’t defined, so there’s nothing to measure them against and no way to compare them over time. Psychologists call this the jingle-jangle problem, where one word means different things and different words mean the same thing. MOSAIC reads for traits defined in published psychology research, like attachment, the Big Five and the pursue–withdraw pattern, each with its own reading guide and its own sources.

  • MOSAIC measures traits defined in published research. A chatbot invents labels on the spot.
  • Each MOSAIC trait has a written definition of what counts and what doesn’t. A chatbot’s labels mean whatever it meant that time.
  • MOSAIC links each finding to the research behind it. A chatbot rarely says where an idea comes from.

SourcesWulff & Mata, 2026·Hazan & Shaver, 1987·Christensen & Heavey, 1990·Driver & Gottman, 2004·Zhu et al., 2025

Pasted into a chatbot

Words that sound right

Describing the chat…No definitions, so nothing to measure.

MOSAIC

Defined traits, with their sources.

Reading for defined traits…Every score measured against published research.

Illustration. The scores are examples.

Your side, or both sides.

Ask a chatbot about your relationship and you set the frame: your question, your selection, your side. And assistants are known to lean towards agreeing with whoever is asking, which researchers call sycophancy. MOSAIC reads both of you with the same method, on the same scale, whoever ordered the report. Messages that cut against a finding count against it, so a pattern has to hold up across the whole history, not just the parts you picked.

  • MOSAIC reads both people with the same method. A chatbot sees only the side you chose to paste.
  • MOSAIC scores you both on the same scale. A chatbot leans towards agreeing with whoever asks.
  • Messages that contradict a finding lower MOSAIC’s confidence score. A chatbot rarely looks for them.

SourcesCheng et al., 2025a·Cheng et al., 2026

Pasted into a chatbot

Your question, your selection

Reading what you chose to paste…Framed by whoever asks, and inclined to agree.

MOSAIC

The same method, for both of you.

Reading both people the same way…One message disagreed, and it counted against the finding.

Illustration.

Accuracy and confidence

How sure is
each finding?

Every finding comes with a confidence level and the messages behind it. Here is what makes one firmer or softer, and why the amount of history matters most.

Evidence10
Against0
ConfidenceModerate
ReadNot readUncertaintyAgainst
  1. 01 · Core

    A snapshot can mislead.

    Core reads a fixed, limited window of recent messages. That is often enough to spot a pattern, but not always enough to be sure of it.

    If the window lands in a tense month, a rough patch can look like the way things always are. Patterns that only surface now and then may not appear at all. So Core findings carry more uncertainty, and more of them read Moderate or Low.

  2. 02 · In-depth

    More history, firmer ground.

    In-depth reads every message, from the first to the latest. The rough patch sits in context with the months around it, and the pattern settles where it really is.

    The uncertainty narrows, more findings reach High, and some findings only exist here: changes that unfold over years, and patterns too rare for a short window to catch.

  3. 03 · Counter-evidence

    The moments that disagree count too.

    Confidence is not only how often something happens. It is also how often it doesn’t. Messages that point the other way are weighed, not hidden.

    A pattern with plenty of exceptions is marked softer to match, and the report says so in plain words instead of rounding it up.

What moves it

Five things decide
how confident it is.

A firm finding has a lot behind it: many moments, over a long time, with few exceptions. You can lean on it. A soft finding is still a real signal, but with less behind it, so give it less weight until you have checked the evidence.

  1. 01

    Volume

    How many separate moments show the pattern.

    More confident when
    It appears in dozens of exchanges.
    Less confident when
    It rests on a handful of messages.
  2. 02

    Span

    How much time those moments cover.

    More confident when
    It holds across months or years.
    Less confident when
    It is packed into one week or one event.
  3. 03

    Consistency

    Whether it repeats in different situations.

    More confident when
    It shows up in calm times and hard ones.
    Less confident when
    It only happens in one kind of moment.
  4. 04

    Both sides

    How much of the exchange is actually in the messages.

    More confident when
    Both people are writing back and forth.
    Less confident when
    One side is quiet, or it moved to calls.
  5. 05

    Exceptions

    How many moments point the other way.

    More confident when
    Counterexamples are rare.
    Less confident when
    The opposite happens nearly as often.

Reading the confidence levels

Low confidence isn’t
wrong. It’s unproven.

Every finding is marked with one of three confidence levels. Here is what each one means for you.

High confidence

Clear, frequent and consistent across time and situations.

Treat it as a solid reading of the pattern.

Moderate confidence

Visible, but with gaps, exceptions or a shorter stretch of history.

Likely true. Check the evidence against what you know.

Low confidence

An early hint from a few moments. Not wrong, just not proven yet.

Worth watching. In-depth often confirms or rules it out.

The honest part

It only knows
what was written.

MOSAIC reads text. However much history it has, some things never reach the page. That is why every finding shows its evidence: so you can check it against everything the messages leave out.

  • Out loud

    Calls, and everything said in person.

  • Tone

    Sarcasm, in-jokes and the look that went with a message.

  • Other chats

    Conversations you didn’t choose, or that happen somewhere else.

  • Your life

    Stress, illness, grief. The context behind a hard month.

It describes patterns in messages. It is not a diagnosis, and it is not the final word. You are.

Getting set up

From download to report.

  1. 1

    Setup

    Download and sign in

    Install MOSAIC on your Mac or phone and create an account. Every report you generate is tied to it.

    • Mac, iPhone and Android
    • One account across devices
    • No message access until you choose
  2. 2

    On your device

    Choose conversations

    Pick exactly which chats you want to understand: one for an Interplay report, or five to ten for Constellation. For Constellation, pick the people who matter most in your life, or for the relationship lens, your past and present partners. On Mac, MOSAIC reads Messages and Contacts with your permission. On mobile, you import a WhatsApp export.

    • You select every conversation
    • One person, or five to ten
    • Relationship Constellation: partners and exes
    • Mac: Full Disk Access for Messages
    • Mobile: WhatsApp chat export
  3. 3

    On your device

    Prepared locally

    Messages are read, contacts are matched and statistics are computed on your device. Phone numbers and email addresses are swapped for anonymous IDs.

    • Your message database never leaves the device
    • Contacts resolved locally
    • Anonymous IDs replace numbers and emails
  4. 4

    Secure analysis

    Analyzed securely

    Only the conversations you selected are sent, over an encrypted connection. The analysis runs as a background job, so you can close the app while it works.

    • HTTPS only
    • Runs in the background
    • Validated before it is saved
  5. 5

    Your report

    Read your report

    Your report is stored against your account and ready whenever you are. It stays private until you choose to share it. Then MOSAIC turns it into a clean, shareable graphic you can send to a friend, a partner or a therapist.

    • Evidence and confidence for every finding
    • Readable on any of your devices
    • Private unless you share it
    • Share as a polished graphic

Get started

Start with one conversation.

Choose a chat that matters to you and see what months or years of messages have been saying all along. Or bring in five or ten, and see the patterns that are yours everywhere.