Vueniverse v0.0.1
Release date: 21 July 2026
Platform: Android preview
Vueniverse v0.0.1 is the first evidence-to-action preview: a private health
timeline that helps people inspect recurring patterns, understand the evidence
behind them, and test small changes without turning AI output into medical
advice.
Highlights
- Explore a complete 30-day Snapshot built from isolated fictional data, or
connect Live sources when permissions and device support are available. - Review heart rate, HRV, sleep, steps, exercise, active calories, calendar
categories, and manual check-ins in a local-first timeline and Observe
dashboard. - Inspect deterministic findings with matched comparisons, effect ranges,
completeness, exclusions, uncertainty, and counterevidence. - Replay repeated event windows with Moment Fingerprint, correct possible
influences, and recompute affected results. - Ask Vueniverse for a bounded MedGemma explanation grounded only in the
evidence already calculated by the app. - Predeclare small personal experiments, track adherence, review the result,
and preserve prior versions in History. - Export a proof receipt containing the result, sources, evidence fingerprint,
analytical version, and the runtime that produced the explanation.
Private and evidence-first by design
- Live and Snapshot records stay in separate encrypted, app-private stores.
- Calendar details are reduced to reviewed category and timing fields; titles,
attendees, descriptions, locations, and organizers are not retained. - The deterministic evidence engine calculates every finding before MedGemma
can explain it. Generated text cannot create, promote, or modify evidence. - Model responses must pass schema, evidence-matching, and safety checks before
they can be shown or cached. Vueniverse uses checked fallback wording when a
model response is unavailable or rejected. - Editing or deleting source data invalidates dependent findings,
explanations, experiments, and exports so stale results are not presented as
current.
On-device MedGemma delivery
Vueniverse does not bundle model weights in the APK. Onboarding downloads the
2.49 GB MedGemma 1.5 4B Q4 model over unmetered Wi-Fi, supports resumable
background transfer, verifies the pinned byte size and SHA-256, and installs the
artifact in app-private storage before Snapshot or Live can continue.
The Android downloader can receive a short-lived signed URL through Firebase
Remote Config, refresh it after an authorization failure, and resume the same
partial artifact. A native Gradle property remains available for controlled
development builds. Credentials and durable signed URLs must never be embedded
in the app.
Known limitations
- This preview targets Android only; iOS is outside the v0.0.1 release scope.
- A valid, authorized model-download URL, sufficient free storage, and an
unmetered connection are required to complete onboarding. The bare private
GCS object URL is not sufficient on its own. - Full real-model delivery acceptance on the API 34 ARM64 emulator and the
physical-phone latency, memory, battery, and thermal benchmark remain release
gates. Until they pass, phone-local MedGemma performance is unverified. - Vueniverse supports personal understanding only. It does not diagnose,
recommend treatment, or replace professional medical care.
Validation and setup
Developers can use the release validation commands
and the journey-ordered manual integration runbook
to verify the app, privacy boundaries, source flows, experiments, exports, and
device-specific MedGemma gates.
Created by @Poujhit @crixalis17
