Lenis

A voice-guided calming companion. Lenis leads you through gentle, EMDR-style bilateral stimulation — a soft target that eases side to side while your eyes follow — and adapts the pace to the sound of your voice, not a slider you drag. It runs entirely in your browser: no account, no server, and your audio never leaves your device.

Lenis is Latin for "soft, gentle, calming."

Built for Hack for Humanity — Summer 2026 (AI for mental & physical health).

Follow the fluffball. It drifts to the pace your voice sets.

It reads your voice	Honest about the evidence	Concussion recovery mode

---

The problem

Self-guided EMDR / bilateral-stimulation apps are everywhere, but they all run on self-report: you tap a 0–10 distress slider and the app adjusts. Two gaps:

1. Nobody reads the body. A number you type is not how activated you actually are. The voice often tells a different story than the slider.
2. No readiness check. Most apps let you jump straight into a distressing memory on screen one — the opposite of how a clinician would pace it.

What Lenis does differently

• 🎙️ Vocal-adaptive pacing. Between sets you speak for a few seconds. Lenis analyses the sound of your voice on-device — pitch variability, speech pace, energy — into a rough arousal proxy, and sets the next set's speed from that. Talk keyed-up, and the target visibly slows down.
• 🧠 On-device coaching, no API key. A free, in-browser transcript (Web Speech API) plus a rule-based coach turns the check-in into a short, warm spoken prompt and a pace decision — and flags crisis language to route you to help. Nothing is sent to a cloud model.
• 🛟 Safety as a feature. Consent gate, always-visible crisis resources, a live "slower / stop" control, "close your eyes and follow the sound" guidance for dizziness, and a self-report-vs-voice comparison so you can see when your words and your body disagree.
• 📊 Honesty on purpose. A built-in "About the evidence" screen lays out, in plain language, what's well-supported (EMDR for PTSD), what's debated (the eye movements themselves), and what's just a proof-of-concept (the voice read).
• 🧠 Concussion recovery mode. The same smooth-pursuit engine powers clinician-aligned oculomotor exercises — smooth pursuit and saccades — with a symptom check before and after and a summary to bring to your clinician. Symptom-titrated: a little challenge is fine, a real flare is your cue to stop.

Responsible-AI stance

Lenis is a self-help calming aid, not therapy and not a medical device. It is not guided trauma processing — full EMDR should be done with a licensed therapist. The vocal signal is a proof-of-concept proxy, not a clinical measure. Everything runs client-side; no audio, transcript, or data is uploaded or persisted. Best used alongside professional care, not instead of it.

How it's built

• React + Vite + TypeScript, single-page, a small state machine (welcome → evidence → setup → session → check-in → close).
• Bilateral-stimulation engine (src/bls.ts): canvas animation with eased motion, stereo-panned Web Audio tones, and haptic taps.
• Voice analysis (src/voice.ts): Pitchy for pitch + Meyda for energy → arousal proxy. Includes a headless, self-testable path (analyzeSamples / synthClip).
• On-device reasoning (src/coach.ts + src/transcribe.ts): free Web Speech transcription + a rule engine. Designed behind a small Guidance interface so a hosted model (e.g. Claude) can be dropped in later without UI changes.
• No backend, no auth, no persistence. Deployed as a static site on Render.

Run it locally

npm install
npm run dev

Then open the local URL. For the microphone features, use Chrome or Edge (Web Speech API); elsewhere the app still runs and simply falls back to the vocal signal without a transcript.

npm run build   # production build → dist/

Credits

Built by patkusch for Hack for Humanity, Summer 2026.