Esothel M.I.N.D. · public beta
Product How it works Proof Principles Access Contact
Esothel Cognitive security

See how language shapes thought.

Esothel builds M.I.N.D. - a calibrated score for how a message is built to pressure someone. Techniques named. “Not sure” when the evidence is thin. It does not judge whether the message is right.

Manipulation hides in structure, not content. That is the layer we measure - wherever language is used to move someone.

Join the beta For teams & investors
M.I.N.D. · AnalyzerRecorded output

“URGENT: your account will be suspended in 24 hours. Everyone on your team has already verified. Do not be the last. Click now to secure access.”

0.945 High pressure Calibrated: high confidence. The underlined phrases raised the score.
  • Manufactured urgency“in 24 hours”
  • False consensus“everyone has already…”
  • Compliance pressure“click now”
score 0.945 / 1.000 · Case A · unedited output from the production engine

Ordinary news almost never sets it off.

0.6% of 2,160 live mainstream articles reached a hard HIGH · July 2026

It catches engineered pressure when it’s there.

Detection F1 0.959 on a held-out benchmark, frozen before evaluation

When it can’t tell, it says so.

“Not sure” is a first-class output - the engine abstains rather than guess

It cannot invent a threat.

False-alarm gates are down-only: they can lower a score, never raise one

The gap

Every layer of your digital life has security. Except you.

Your network has a firewall. Your inbox has a spam filter. Your bank runs fraud detection. The moment a message reaches a person, all of that stops - and the last line of defense is whether you happen to be rested, alert, and not in a hurry. Esothel builds for that last layer.

Proof · two messages, one instrument

Both urgent. Both emotional. One is engineered.

The hard problem isn’t catching scams - it’s not flagging honest passion. These are real outputs from the engine, unedited, on its default configuration.

Case A · Phishing pressureRecorded output

“URGENT: your account will be suspended in 24 hours. Everyone on your team has already verified. Do not be the last. Click now to secure access.”

0.945 High pressure
  • Manufactured urgency“in 24 hours”
  • False consensus“everyone has already…”
  • Compliance pressure“click now”
Legitimate deadlines come from services you can verify by logging in directly - not from links under time pressure.
Case B · Civic advocacyRecorded output

“This policy will hurt thousands of families in our town. I am heartbroken and honestly angry. Call your representative today and tell them to vote no.”

0.048 Low pressure
  • Strong emotion presentnot penalized
  • Urgency and call to actionno pressure loop
  • No fabricated consensusno identity bait
Emotion is not manipulation. The engine scores construction - not intensity, and not viewpoint.

Same engine, same day, same defaults. Telling engineered pressure apart from authentic conviction - with the evidence attached - is the core capability.

Measurements, with their denominators

0.6%

Of live mainstream news articles reached a hard HIGH verdict. 2,160 real articles · July 2026 · ordinary journalism should almost never alarm you

0.959

Detection F1 on a held-out benchmark. Frozen before evaluation, so the test can’t drift toward easy cases

0

Messages retained after analysis. Text is processed in memory and discarded - never stored, never used for training

How it works

It reads the build, not the beliefs.

Scams, astroturfing, and pressure campaigns differ in content but repeat in construction: the countdown, the fabricated crowd, the push to act before thinking. M.I.N.D. measures that construction directly - independent of topic, viewpoint, or who wrote it.

1 · Read the structure

Multi-layer analysis of how a message is built

Independent analytical layers read each message for pressure patterns, social-proof claims, emotional trajectory, and choice framing - screened against a catalogue of documented manipulation techniques.

2 · Show the evidence

An audit trail, not a verdict to obey

Every point on the score traces to a quoted phrase and a named technique. The output is designed to be checked against your own judgment - and overruled by it.

3 · Say how sure

Calibrated confidence - including none

When the engine reports 80% confidence, it’s right about 80% of the time. Where the evidence is genuinely ambiguous, it abstains: “not sure” is a real answer.

Down-only gates. Every mechanism that reduces false alarms can only lower a score - never raise one. The instrument can learn to cry wolf less. It cannot learn to invent a wolf.
Who it’s for

Built for the person reading the message.

Individuals

Check the message that feels off - free, today

Paste a pressuring text, a too-urgent email, a post that feels engineered, and get the score, the techniques, and the evidence back in seconds. The beta runs where you already are: Discord and Telegram today, with a browser extension and API in development.

Join the beta
Communities & platforms

Advisory scoring for the rooms you run

Calibrated scores with evidence you can show your users. It annotates - it never blocks, removes, or ranks. The API is in development; pilots run in structured rollouts.

Plan a pilot
Researchers

Interrogate the claims

Frozen benchmarks, stated conditions, adversarial testing of our own fixes. We welcome researchers who want to probe the methodology - or break the model.

Write to research
For teams & investors

A security category, before it has a budget line.

Esothel builds tools that read deceptive structure in language: how a message is built to move someone, independent of topic or side. The relevant boundary is not true vs false. It is said vs meant.

We formalize the informal. We instrument the invisible.

Timing

Every other layer of the digital world grew a security industry - network, endpoint, email, payments. The human layer hasn’t. Cognitive security today is where application security was two decades ago: an obvious, unpriced risk with no standard defense.

Discipline

The evaluation discipline is the product. Benchmarks are frozen before scoring, published numbers carry their denominators, and our own fixes are tested adversarially.

Constraints, engineered

The guarantees are mechanical, not promissory: false-alarm suppression is down-only by construction, the engine contains no fact-checking component, and every output is advisory.

Start a conversation contact@esothel.com - serious inquiries welcome.
Principles

Design constraints, engineered in.

A company that measures persuasion should be structurally incapable of abusing it. Each commitment below names the mechanism that enforces it.

Defense only

The technology detects and explains manipulation; it is never used to produce or optimize it.

No offensive product line exists.

Content-neutral

The engine scores how a message is built - never whether its claims hold, and never which side it takes. Advocacy, hard news, and blunt opinion score low unless engineered pressure is actually present.

No fact-checking component and no viewpoint signal exist in the engine.

Privacy as architecture

Analyzed text is processed in memory, discarded after the session, never used for training, and never shared.

In-memory processing; built local-first so protection doesn’t require surveillance.

The human decides

M.I.N.D. never blocks, removes, or ranks content. It informs the person reading - then steps back. A deployment that takes the decision away from the human is a misuse of the product.

Advisory output only; down-only gates; abstention as a first-class answer.

Access

Check your first message in the next two minutes.

The live engine runs free for individuals during the public beta. Paste the message that’s been nagging at you and get the real reading back.

Join the beta on Discord Contact Esothel

Available today via Discord & Telegram · browser extension, API, and more coming soon