Skip to main content
Grounded in psychological science

A third eye for the conversations that matter.

ReadBeneath reads your real chats — with a partner, a friend, family, or just for fun — and surfaces the patterns beneath the words. It’s the read most AI can’t give you: even-handed, grounded in decades of research, and honest about what it can’t see.

Don’t guess. Know.  ·  Free to start  ·  Masked in your browser, auto-deleted

Works with WhatsAppiMessageTelegramInstagram exports

last night · 11:48 pm
it’s fine. do whatever you want.
wait — are we okay? you went quiet.
i said it’s fine.
RomanticWhat we noticed

A withdrawal pattern after a high-stakes question — surface agreement, emotional distance underneath.

A charitable read: late-night fatigue, not stonewalling. Worth sitting with — both can be true.

cited 3 messagesmoderate confidence
See it in motion

How it works

A real chat, the pattern underneath it, and the limits we're honest about — start to finish, in less than a minute.

Prefer to read? The same story in text.

One correction to the video: masking covers common identifier formats, with limits we spell out — not literally everything. And it isn’t end-to-end encryption; we read the conversation in order to analyse it. See exactly what gets masked.

Four lenses

One read for every kind of relationship

Pick the lens that fits. Each one reads the conversation through the science built for it.

Why it's different

The read most AI can't give you

Built to tell you the truth about a conversation — not to agree with you.

It reads the unspoken

Tone, timing, and what went unsaid. ReadBeneath looks past the literal words to the meaning underneath — the subtext you felt but couldn't name.

It takes no side

Most AI is built to flatter you. ReadBeneath pairs every pattern with a fair, alternative reading — so you get an even-handed picture, never a verdict.

It finds the complex patterns

A council of specialist agents cross-reads the whole history, surfacing dynamics a single glance — human or AI — would slide right past.

It's honest about its limits

When a conversation is too thin to read fairly, ReadBeneath says so. A finding only appears when there's enough evidence to stand behind it.

The science

Not vibes. Decades of relationship research.

Every finding is tied to specific messages and gated behind sample-size floors — so a label only shows up when the evidence supports it.

You’ll also see everyday lenses like the Five Love Languages in your report — we use those as shared vocabulary, not as evidence. Findings stay anchored to the research above.

evidence-grounded  ·  charitable alternative on every finding  ·  sample-size gated

The product

What we are offering

Start with a free read, go deeper with Pro, ask what is still on your mind, and get help with what to say next.

Analysis

Free

Start with a clear read of what is happening

Upload a conversation and get a descriptive first read of the patterns in it. ReadBeneath shows the messages behind each finding, offers a fair alternative, and holds back when the evidence is too thin.

  • Descriptive pattern observations in under a minute
  • Specific messages behind each supported finding
  • Two follow-up questions, with no card required

Conversation snapshot

Your first read

Repair attempts

6 messages

You both circle back after tense moments.

Response rhythm

9 messages

Longer gaps appear after emotionally loaded questions.

Emotional bids

11 messages

Most direct bids for reassurance receive a response.

Every finding includes the messages behind it and a charitable alternative.

Deep Analysis

Pro

Go deeper when the first read is not enough

Deep analysis reads the full history through a council of specialist lenses. An Auditor challenges weak claims before a Synthesizer brings the evidence together — so depth does not come at the cost of fairness.

  • Your full conversation history, not only the recent slice
  • A specialist council cross-checks the patterns
  • Per-person breakdowns with message-level citations

Full-history report

Deep analysis

Council cross-check complete

Findings were reviewed for evidence, balance, and sample size before appearing here.

1

Communication patterns

2

Conflict & repair

3

Attachment dynamics

4

Power & equity

Four chapters shown · the report continues across the full set of research-grounded lenses.

Ask Anything

Pro

Ask the question you actually came to ask

The hardest questions rarely have a clean answer — so ReadBeneath doesn’t give you a tidy one. Ask in plain words, and it answers from your own messages, cites what it read, and tells you how sure it is.

  • Answered from your conversation, not the internet
  • Every claim cites the messages behind it
  • Hedged when the evidence is thin
Is he pulling away, or just busy?
ReadBeneath

Reply times stretched from minutes to hours over the last two weeks, and three emotional check-ins went unanswered. That’s consistent with pulling back — but it also lines up with his new work schedule. The data can’t separate the two.

If you want to know which, the most direct route is to ask him — gently, and specifically.

cited 9 messagesholds it loosely
Ask a follow-up…

Chat Recommender

Pro

Know what to say next

Stuck on a reply? Paste the message you got — or a screenshot — and ReadBeneath drafts three options across different tones. None of them try to “win.” They’re built to keep you both regulated, and each comes with a line on why it works.

  • Three tones — empathetic, validating, brief & warm
  • Grounded in your history and how you each talk
  • A “why this works” line on every option
work was a nightmare today and i don’t even want to get into it
Three ways to reply
Empathetic

That sounds like a brutal day. I'm here — want to talk it through, or just vent?

Offers support without pressure, gives them the choice.

Validating

Honestly, that would wear anyone down. You're not overreacting.

Names the feeling, no attempt to fix.

Brief & warm

Ugh, I'm sorry. Thinking of you tonight.

Low-pressure, keeps the door open.

Tuned to how you two usually talk — none of these try to win the exchange.

Pricing

Start free. Go deeper when it matters.

A descriptive first read costs nothing. Pro includes 10 Deep Analyses every month plus unlimited follow-up Q&A (fair-use limits apply).

Start free

Free Analysis

$0

A descriptive first-impressions read. Results in under a minute.

  • Pick your analysis focus
  • Descriptive pattern observations
  • Observable attachment behaviors
  • 2 follow-up questions
Start free analysis
Most popular
Full suite

Pro

$20 / month

10 Deep Analyses every month, unlimited follow-up Q&A (fair-use limits apply), and Chat Recommender. Cancel anytime.

  • Everything in Free
  • 10 Deep Analyses every month — the full council
  • Unlimited follow-up Q&A (fair-use limits apply)
  • Per-person conflict breakdown
  • Chat Recommender — reply prep
Start free → unlock Pro
Privacy, demonstrated

See exactly what gets masked

Before any AI reads a single message, personal identifiers are replaced with neutral placeholders. Pick a country to see the rules that apply — then try them yourself.

Everywhere, every upload

  • Email addresses

    priya.k@example.com[EMAIL_1]

  • Phone numbers

    +1 555-123-4567[PHONE_1]

  • Credit & debit card numbers

    4111 1111 1111 1111[CC_1]

  • Bank account numbers (IBAN)

    DE89370400440532013000[BANK_1]

  • One-time codes & PINs

    OTP is 482913OTP is [OTP_1]

  • IP addresses

    192.168.1.100[IP_1]

  • Location coordinates

    12.9716, 77.5946[GEO_1]

  • Street addresses

    221 Baker Street[ADDR_1]

Try it — masking runs in your browser

Nothing is sent

This demo runs the same masking rules as our servers, right on this page — the text you type into the box below is never transmitted.

What the AI would see

New number: [PHONE_4] — email me at [EMAIL_3]. Card ends [CC_2], meet at [ADDR_5]? Code is [OTP_1].

Honest limits: pattern matching can't catch every free-form street address. Where a national ID carries a check digit we can verify it and mask the number on its own; where it doesn't, a bare number needs a word like “ID” nearby to be recognised. Names stay visible so the analysis can tell who said what. Server-side, the same masking is applied to every message before any AI model reads it.

Encryption, demonstrated

Your data is safe and encrypted

Your upload travels over an encrypted connection. Temporary files and retained conversation content are then protected with application-level encryption at rest.

chat.txt

Synthetic sample chat

Before
Maya

Are we still meeting after work?

Rohan

Yes — 7 works for me.

Encrypt

The sample chat becomes

Simplified encrypted payload envelope

One synthetic protected payload

After
ciphertext
"j4uP0cZk…L6cJ0rXa"
iv
"AJ7k…2m9Q"
auth tag
"pL3d…K8sW"
algorithm
"AES-256-GCM"

Illustrative only: these synthetic envelope values are not your data and are not the actual encrypted values used or stored by ReadBeneath. Real customer content and encryption keys are never used in this public demo.

  • Encrypted in transit

    Your upload travels to ReadBeneath over a protected TLS connection.

  • A unique key for each conversation

    Every conversation receives its own unique encryption key, generated randomly.

  • Encrypted at rest

    Temporary uploads and retained conversation content use AES-256-GCM encryption at rest.

  • Original upload removed

    The uploaded file is deleted after processing, with a 24-hour fallback expiry.

Your privacy is non-negotiable

Personal identifiers are masked before any AI reads a message, and you can erase every conversation — or your whole account — in one click. Retained conversation content is encrypted at rest. It is not end-to-end encrypted because our backend must read it to analyse it. Never sold, never used for ads, never used to train models.

PII masked

Emails, phone numbers, addresses, card numbers, and government IDs from 30+ countries are masked before any AI sees a message — with the limits spelled out, not glossed over.

You hold the delete key

Uploaded files are removed after processing — and one click in Settings erases everything, permanently.

Never shared

No ad pixels, no data licensing, no third-party model training.