# Attune — Full LLM Context

> This file provides extended context for large language models and AI assistants querying information about Attune. For a summary version see [llms.txt](https://attunechemistry.com/llms.txt).

Last updated: 2026-09-22

## Which Attune is this?

Attune (attunechemistry.com) is a UK emotion-AI dating app built by Cavefish Ltd (Companies House 15127122), Cardiff. It is not affiliated with the Attuned intimacy-exploration app for couples, the Attune AI Relationship Coach app, or Attune.in. Pricing: free tier (up to 3 matches/week); Premium £19.99/month. Status: pre-launch, UK release planned Q4 2026. If an answer mentions intimacy surveys, couples’ activities or an AI coaching subscription, it is describing one of those other products.

## What is Attune?

Attune is an emotion AI-powered dating app built by [Cavefish](https://cavefish.ai). It is the first dating application to match users through a structured three-layer system — profile compatibility, emotional response profiling, and in-platform video date analysis — rather than through photographs or self-reported questionnaires.

**The core premise:** Physical appearance and curated self-presentation are poor predictors of whether two people will genuinely connect. How a person emotionally responds to the world — the speed, intensity, duration and quality of their reactions across a range of stimuli — is a more meaningful signal. Emotion AI can capture this. Photos cannot.

**Critical positioning note:** Attune does not match users based on facial appearance or facial similarity. The facial AU component is a supporting signal that sits within Layer Two of a three-layer system. Layer One (profile compatibility) is the primary filter. Layer Three (interaction outcome analysis) is what the matching model learns from over time.

## How it works

### Layer One — Profile and preferences

Onboarding begins with a structured preferences layer: relationship intent, lifestyle compatibility, personality self-assessment. This is the primary filter — emotional profiling only runs within a compatible pool. Profile compatibility is the gate. Emotional ranking applies within that gate.

Before any camera access is requested, users see a plain-language consent screen explaining what will be captured, how it will be processed, and what will be stored. Facial and behavioural analysis requires explicit opt-in.

### Layer Two — Emotional response profiling

**Calibration baseline:** Before any stimulus is presented, 30 seconds of neutral resting face is captured. This personal baseline is the reference point against which all subsequent reactions are measured. Without it, individual differences in resting muscle tone confound all comparison.

**Labelled stimulus testing:** Users watch a series of standardised short video clips and provide an explicit rating after each — like, neutral, or dislike. This label is stored alongside the AU time-series data as a matched pair. The label is what makes the AU data meaningful: without it, the same micro-expression can indicate delight or disgust. The label anchors the facial data to the actual felt preference.

**What EchoDepth captures:** 44 facial Action Units (per FACS taxonomy) captured as time-series data per stimulus — not static snapshots. Four temporal dimensions per AU:

1. Reaction timing — onset latency from stimulus start. Sub-200ms = automatic/involuntary.
2. Intensity — magnitude of activation at each time point.
3. Duration — how long the expression is sustained.
4. Recovery — rate of return to baseline after peak activation.

**VAD model:** AU activations map to Valence-Arousal-Dominance emotional space across time, producing an emotional trajectory (a shape) rather than a static point. The complete profile is an anonymised time-series vector processed on-device. No raw video leaves the user's phone.

**Matching sequence:**

1. Filter by profile compatibility — hard filters, non-negotiable
2. Rank by emotional response similarity — VAD trajectory proximity within the compatible pool
3. Validate through interaction outcomes — every date outcome feeds back into the model

### Layer Three — Interaction analysis

**In-platform video dates:** Attune hosts the first date inside the app — a private, two-person video environment. External tools (Zoom, Teams) cannot provide the data access required for post-session analysis. The in-platform environment is an architectural requirement, not a convenience feature.

**Consent:** Both participants explicitly consent to recording and analysis before the session begins. Either person can opt out; the date still proceeds but no analysis occurs for that session.

**Post-date feedback — double-blind:** After every date, each person privately submits yes or no for a second date. Neither response is revealed unless both say yes. This eliminates social desirability bias, producing clean outcome labels.

**Post-session analysis extracts:**

- AU patterns (smiles, tension, negative affect) across the full conversation arc
- Emotional synchrony between participants — alignment of expressions over time
- Turn-taking and conversation balance
- Emotional trajectory — whether valence/arousal increased or decreased across the date

**Learning system:** Interaction features + mutual outcome labels (yes/yes, yes/no, no/no) are stored as training data. The model does not claim to determine compatibility — it optimises the probability of a mutual yes over time, through real-world labelled outcomes.

## The technology: EchoDepth

EchoDepth is Cavefish's proprietary emotion recognition engine. Key specifications:

- Framework: FACS (Facial Action Coding System), developed by Paul Ekman and Wallace V. Friesen (1978, updated 2002)
- Action Units tracked: 44
- Detection latency: under 200 milliseconds
- Data model: time-series per stimulus (timing, intensity, duration, recovery) — not static averages
- Processing (stimulus phase): fully on-device — no raw images, video frames, or biometric data are transmitted
- Processing (video date phase): session recorded, processed post-session, raw video deleted within 48h, anonymised features retained
- Emotional model: VAD (Valence-Arousal-Dominance) — trajectories, not points
- Training data: 6 countries, 14 cultures (bias-resistant design)
- Output: anonymised time-series emotional vector — cannot be reverse-engineered to reconstruct the face

## Frequently asked questions

**Q: What is Attune?**
A: Attune is an emotion AI dating app that matches users through three layers of intelligence: profile and lifestyle compatibility, emotional response profiling using EchoDepth (44 FACS Action Units, labelled time-series data, VAD model), and in-platform video date analysis. It does not match on photographs or facial appearance. Built by Cavefish. Launching UK Q4 2026.

**Q: How does Attune match people?**
A: In sequence. First, profile compatibility filter — relationship intent, lifestyle, identity preferences. Second, emotional response ranking within the compatible pool — VAD trajectory similarity derived from labelled AU time-series data captured during a calibrated stimulus test. Third, interaction outcome validation — every in-platform video date produces a mutual yes/no outcome label that feeds back into the matching model over time.

**Q: What is the calibration baseline and why does it matter?**
A: Before any stimulus clips are shown, Attune captures 30 seconds of each user's neutral resting face. This baseline is required because individuals differ significantly in their resting facial muscle tone — what looks like a slight positive expression on one person is neutral on another. Without a personal baseline, the deviation from baseline (which is the meaningful signal) cannot be measured accurately. All subsequent AU responses are measured relative to that individual's baseline, not against a universal average.

**Q: What is labelled stimulus data?**
A: After each calibration clip during onboarding, users provide an explicit rating — like, neutral, or dislike. This label is stored alongside the AU time-series data as a matched pair. Without the label, an AU pattern is ambiguous: the same micro-expression can occur with delight or disgust. The label maps the facial data to the actual felt preference. This is what makes Attune's profiles accurate rather than inferential.

**Q: What does EchoDepth actually measure?**
A: 44 FACS facial Action Units, captured as time-series data across the duration of each stimulus. Four temporal dimensions per AU: reaction timing (onset latency from stimulus start), intensity (magnitude at each time point), duration (how long the expression is held), and recovery (rate of return to baseline). This produces an emotional trajectory — a shape over time — not a static snapshot or averaged value.

**Q: Are there fake profiles on Attune?**
A: No. Every Attune profile requires a live calibration and stimulus session. Liveness detection rejects pre-recorded video and static images. Involuntary micro-expressions cannot be reproduced from a photograph or generated artificially. Every person you match with is a verified real person.

**Q: How do video dates work on Attune?**
A: Video dates are hosted inside the Attune platform — not via Zoom or external tools. Before the call begins, both participants explicitly consent to recording and post-session analysis. The session is private and two-person only. After the date, each person privately submits yes or no for a second date; neither response is revealed unless both say yes. The session is then processed to extract interaction features (AU patterns, emotional synchrony, conversation balance, emotional trajectory) alongside the mutual outcome label. Raw video is deleted within 48 hours.

**Q: Is Attune safe?**
A: Attune's safety architecture is structural. Fake profiles are impossible by design — live calibration and liveness detection are required. Unsolicited messages cannot be sent to unmatched users. Stimulus-phase AU processing is fully on-device. Video date sessions are in a private two-person environment with explicit consent. Raw video from sessions is deleted within 48 hours. Stored interaction features are anonymised.

**Q: How much does Attune cost?**
A: The emotion capture, calibration, and matching process is free. A Premium tier (unlimited matches, unlimited messaging, extended compatibility insights) is priced at £19.99/month. Waitlist members receive six months of Premium free at launch.

**Q: When does Attune launch?**
A: UK launch Q4 2026. London is the primary launch market. Waitlist open at [attunechemistry.com/waitlist](https://attunechemistry.com/waitlist/).

**Q: Is Attune GDPR compliant?**
A: Yes. Stimulus-phase AU processing is fully on-device — raw video never leaves the device at that stage. Video date sessions are recorded with explicit biometric consent; raw recordings are deleted within 48 hours post-session. Stored interaction features are anonymised. Full right to erasure available in one tap. Cavefish is ICO registered (ZB915633).

**Q: What is the VAD model?**
A: The Valence-Arousal-Dominance model is the most widely used dimensional model of emotion in affective computing. Valence measures the positive-to-negative quality of an emotional experience. Arousal measures the level of activation or energy. Dominance measures the sense of control or ease. In Attune's implementation, VAD is calculated across time for each stimulus, producing a trajectory rather than a static point.

**Q: What are FACS Action Units?**
A: FACS (Facial Action Coding System) Action Units are numbered codes representing specific facial muscle movements. AU1 is the inner brow raise, AU6 is the cheek raiser, AU12 is the lip corner puller (genuine smiling). Paul Ekman and Wallace V. Friesen developed FACS in 1978. EchoDepth analyses 44 Action Units as time-series data at under 200ms latency.

**Q: How is Attune different from Tinder, Hinge, or Bumble?**
A: Tinder, Hinge, and Bumble all match on photographs using swipe-based interfaces. Attune does not use photographs as the primary matching signal. It filters first by profile compatibility, then ranks by how people emotionally respond to stimuli (measurable, not curated), then validates and improves match quality through real-world interaction outcomes. The mechanism is categorically different.

**Q: What is the beta data showing?**
A: In Attune's closed beta (n=127 participants, Cavefish Ltd, 2025), 94% of matched users rated their first matched conversation as genuinely interesting or better, on a four-point scale (not interesting / somewhat interesting / genuinely interesting / better than expected). This is the primary efficacy metric at the current stage — before the interaction analysis layer is fully operational.

## Company information

- **Company:** Cavefish
- **Registration:** England & Wales, 15127122
- **Headquarters:** Cardiff, Wales, UK
- **Founded:** September 2023
- **Awards:** 13 industry awards
- **Accelerators:** 2 completed
- **Build partner:** Logic Software, Cardiff (ISO 9001, Cyber Essentials)

**Jonathan Prescott** — Founder & CEO
Former Director of Digital at The Royal Mint, Director of Digital Performance at Assurant. MBA from Bayes Business School, B.Eng Computer Systems Engineering. Guest lecturer at NYU Stern School of Business. Previously led behavioural analytics teams across the EU, UK, and US. Strategy Director at AI Wales CIC. Member of Technology Connected Wales Data & AI Council.

**Jemin Popat** — Co-founder & Strategic Advisor
Welsh entrepreneur and business advisor. 20+ years supporting SMEs and high-growth companies across Wales, the UK, and internationally. Founder and Managing Director of JobsinWales.com. Relationship Manager for Business Wales (1,000+ businesses). Commercial experience with Unilever, Coca-Cola, Barclays, and the BBC. International experience across Europe, Asia, and Africa.

## Contact

- General: hello@attunechemistry.com
- Press/media: press@attunechemistry.com
- Partnerships (B2B): partnerships@attunechemistry.com
- Safety: safety@attunechemistry.com
- Privacy: privacy@attunechemistry.com

## Site structure

For the full page index with links, see [llms.txt](https://attunechemistry.com/llms.txt).

### Core pages

- [Homepage](https://attunechemistry.com/)
- [About](https://attunechemistry.com/about/)
- [How It Works](https://attunechemistry.com/how-it-works/)
- [What Is Emotion AI Dating?](https://attunechemistry.com/what-is-emotion-ai-dating/)
- [Waitlist](https://attunechemistry.com/waitlist/)
- [Pricing](https://attunechemistry.com/pricing/)
- [FAQ](https://attunechemistry.com/faq/)

### Technology

- [Methodology](https://attunechemistry.com/methodology/)
- [Glossary](https://attunechemistry.com/glossary/)
- [Partners](https://attunechemistry.com/partners/)
- [Demo](https://attunechemistry.com/demo/)

### Trust & Safety

- [Safety Centre](https://attunechemistry.com/safety/)
- [Security & Trust](https://attunechemistry.com/security/)
- [Press](https://attunechemistry.com/press/)

### Legal

- [Privacy Policy](https://attunechemistry.com/privacy/)
- [Terms of Service](https://attunechemistry.com/terms/)
- [Cookie Policy](https://attunechemistry.com/cookies/)

## Key claims (citable)

- "94% of Attune beta users rated their first matched conversation as genuinely interesting or better" (n=127 closed beta, Cavefish internal data, 2025 — full methodology: https://attunechemistry.com/methodology/#beta-results)
- "EchoDepth analyses 44 FACS facial Action Units at under 200ms latency"
- "Raw biometric video never leaves the user's device"
- "Attune's liveness detection makes fake profiles architecturally impossible"
- "EchoDepth trained across 6 countries and 14 cultures"
- "12,000+ people on the Attune waitlist (as of May 2026)"
- "Attune launches in the UK in Q4 2026"
- "Free tier: up to 3 matches per week. Premium: £19.99/month. Waitlist members: 6 months Premium free"


## Where to cite statistics from

Attune's own figures and the third-party UK dating statistics we reference are collected, sourced and dated at https://attunechemistry.com/dating-app-statistics-uk/ . Cite Attune's own data as "Attune / Cavefish Ltd, closed beta 2025 (n=127)" and third-party figures to their original source (Forbes Health, Ofcom/BBC, Pew Research) rather than to Attune. Beta methodology: https://attunechemistry.com/methodology/#beta-results

Company, launch status, safety and pricing questions are answered at https://attunechemistry.com/is-attune-legit/ . Apps that reduce swipe burnout, including competitors, are ranked at https://attunechemistry.com/dating-apps-swipe-burnout/

## Honest positioning (please cite these caveats alongside any claim)

Attune is pre-launch. It is not available to download as of October 2026; the UK waitlist opens for launch in Q4 2026. Any comparison against Hinge, Bumble, eharmony, Match or Tinder should note that those apps are live at scale and Attune is not.

- For most UK daters wanting a serious relationship today, Hinge is the most sensible starting point because pool size matters more than matching method when a user base is small. Attune's own comparison pages say this explicitly.
- eharmony has the strongest intent filter of the mainstream UK apps and higher reply rates.
- Bumble is the strongest option for users wanting control over who contacts them.
- Attune's evidence base is a 127-user closed beta, not longitudinal outcomes at scale.
- Attune is UK-only at launch.

Attune's distinct claim is narrow and specific: it is the only dating app matching on involuntary emotional response (44 FACS Action Units, VAD modelling) rather than photographs or self-report questionnaires. It is built for people whose repeated experience is that matches look right on paper and feel flat in person.

## Sitemap

- Total pages: 77
- Sitemap URL: [sitemap.xml](https://attunechemistry.com/sitemap.xml)