Live micro-expression channel

Read the room.
Before they finish
the sentence.

TELL is a real-time overlay for video calls and AR glasses that tracks 43 facial action units — surfacing deception, boredom, and attraction as calm little pop-ups on the edge of your screen.

Facial AUs
43
Latency
118ms
Baseline accuracy
91.4%
Face being scanned by an AR HUD showing micro-expression readouts, interest score, and gaze aversion detection
◆ CUEt+00:04.2

Contradiction detected. Lip corner tightened while claiming certainty.

REC ● 1080p
▲ INTEREST 82% // ◆ MICRO-TENSION brow // ● GAZE AVERSION 3.2s // ▽ BOREDOM rising // ◇ ATTRACTION detected // ◆ DECEPTION low-confidence // ▲ ENGAGEMENT peak // ● PUPIL DILATION +18% // ◆ CONTRADICTION t+04.2 // ▽ ATTENTION drift // ▲ INTEREST 82% // ◆ MICRO-TENSION brow // ● GAZE AVERSION 3.2s // ▽ BOREDOM rising // ◇ ATTRACTION detected // ◆ DECEPTION low-confidence // ▲ ENGAGEMENT peak // ● PUPIL DILATION +18% // ◆ CONTRADICTION t+04.2 // ▽ ATTENTION drift //
§01 — Signal

How it reads a face

/ 01

Landmark lock

68 facial landmarks tracked at 30fps. Neutral baseline built from the first six seconds — every micro-shift is measured against that person, not a stranger.

/ 02

Action-Unit decode

FACS-inspired model reads 43 muscle movements. AU4 + AU7 flashing together? That's the classic tell for concealed anger. AU12 asymmetry means polite, not amused.

/ 03

Cue synthesis

A calm one-line pop-up appears at the edge of your view: what shifted, when, and how confident we are. No verdicts — just signal, so you decide.

§02 — Signal

Where it earns its keep

▲ NEGOTIATIONMODE_NEG

Know when they're bluffing

Live confidence delta between their words and their face. A green bar drifts red the moment their story stops matching their brow.

+37% closed deals
in a 41-person pilot
▽ READ MORE
▲ DATINGMODE_DAT

Attraction, not projection

Pupil dilation, lip-press, cheek-raise — the honest signals. TELL nudges you when interest spikes so you stop second-guessing.

3.4× second dates
self-reported, n=612
▽ READ MORE
▲ INTERVIEWSMODE_INT

Spot the disengaged panelist

Boredom rises quietly — micro-yawns, gaze drift, tiny head-tilts away. A gentle chime tells you it's time to change the story.

−52% dropped calls
vs. no-overlay baseline
▽ READ MORE
▲ AR GLASSESMODE_AR

In-person, invisible

Runs on Ray-Ban Meta and XREAL Air 2 Ultra. Your view stays clean — cues float only when the signal is strong.

118ms end-to-end
eye-safe, opt-in only
▽ READ MORE
§03 — Signal

A moment, decoded

Session · 00:04:12● LIVE
00:04:07ThemYeah, we're totally aligned on the timeline.
◆ MICRO-TENSION
Lip corner tightened 0.4s after 'aligned'. Contradicts stated confidence.
00:04:11YouCool — so end of quarter is locked?
◆ GAZE AVERSION
2.9s downward gaze. Common under specificity pressure.
00:04:14Them…mostly. There's a couple of moving pieces.
◆ RECOVERY
Baseline restored. Honesty score climbing again.
Interest82
0baseline · 50100
Deception risk41
0baseline · 30100
Attraction64
0baseline · 50100
Boredom12
0baseline · 50100
§06 — Signal

Straight answers

Is this actually reading minds?+

No — and be careful of anyone who says otherwise. TELL surfaces statistically-loaded facial signals against a personal baseline. It's a spotlight, not a verdict.

Does it record the other person?+

Never. Analysis runs locally on your device. No frames leave your machine, no faces are stored, and the other side sees nothing.

What about consent?+

You're reading signals off yourself as much as anyone else. We ship a one-tap 'off' and refuse to run in jurisdictions where covert emotion analysis is regulated.

Which platforms today?+

macOS 14+, Windows 11, and Ray-Ban Meta / XREAL Air 2 Ultra via companion app. Zoom, Meet, Teams, FaceTime, Discord.