Attune keeps a message's real meaning intact while translating only the framing — in both directions — and it helps put words to feelings when the feelings themselves are hard to locate. Runs on a multi-model Gemini 3 pipeline through Puter.js: no API key, no backend, nothing to sign up for except a free Puter account the first time you use it.
This isn't one prompt in, one string out. A fast model drafts three phrasing candidates at nearby directness levels; a stronger reasoning model scores each for how well it preserves your meaning versus how well it fits the intended tone, then picks and explains its choice.
Built for alexithymia. Gemini 3 Pro reasons about your sensations twice — once weighting the body signals, once weighting the situation — and only labels an emotion "cross-validated" if both passes agree.
Based on Social Stories — a well-established, evidence-informed technique for previewing an unfamiliar or anxiety-inducing situation with calm, descriptive, first-person framing. Gemini 3 Pro writes a 4-step script; Gemini's image model illustrates each step; Gemini TTS reads the whole thing aloud.
A judgment-free partner to try a hard conversation on, at your own pace, with feedback after every line. One model plays the other person; a second model — running independently, never shown to the "partner" — coaches your side in real time.
Turn "I don't know how to ask for this" into a message you can actually send — a short version for a text or Slack message, a longer one for an email — without flattening it into generic corporate ask-speak.
Two taps. No streaks, no guilt copy. Just a record you can point to when you need to explain "I'm at capacity" without having to explain it in words. Deliberately not AI — not every feature needs a model.
No AI here — plain statistics computed live from what you've logged in this session: a rolling battery trend, a same-session anomaly check, and a simple time-window correlation between battery dips and confirmed emotions. Nothing is inferred beyond what the numbers support.
Nothing here is a single prompt wrapped in a text box. Each feature routes work to the model shaped for that job, through Puter.js — free, no API key, no backend to run.
Generates 3 candidate phrasings per translation in the Bridge. Cheap and quick, because it's disposable — most candidates get discarded.
Scores Bridge candidates, runs the two-pass emotion cross-validation, writes the Social Story script, coaches each rehearsal line independently of the roleplay partner, and drafts self-advocacy notes.
Illustrates each Social Story panel and narrates the Bridge output and story aloud, for anyone who reads better by ear or needs the reading-fatigue relief.
Insights runs on no model at all — it's plain statistics (rolling averages, a z-score anomaly check, a time-window correlation) computed live from what you've actually logged this session, deliberately kept free of AI inference so it never claims more than the numbers support. Everything else degrades gracefully: if Puter or the model call fails or is declined, each feature falls back to transparent, rule-based local logic (see index.html) so the app is always demoable, and the status pill always tells you which mode is active.