Arthur

## Inspiration

Every personal AI assistant on the market suffers from the same critical flaw: it is tethered to distant cloud servers. When you are in an airplane, in a dead zone, or simply valuing your privacy, your assistant dies or
risks leaking your personal life to third-party data centers.
We wanted to prove the opposite: that an AI assistant can be 100% sovereign, local-first, answer in sub-
millisecond latency, and sustain an honest business model without ever compromising user privacy.

## What it does

Arthur Mobile is a zero-latency, local-first personal AI assistant running entirely on the smartphone.

• 100% Local & Sovereign: It embeds a compressed knowledge brain (230 topics, 192 KB) inside the device. It
requires 0 network calls to answer.
• Instantaneous: It responds in 0.13 ms in airplane mode.
• Intellectual Honesty: When Arthur does not know the answer, he explicitly says "I don't know" rather than
fabricating hallucinations.
• Robust RevenueCat Monetization: A deterministic state machine controls access:
• FREE_TIER: 3 free questions per day.
• PREMIUM_ACTIVE: Unlimited queries verified through the RevenueCat SDK.
• OFFLINE_GRACE_PERIOD: A 72-hour sliding grace window so subscribers traveling without internet never get locked out.

## How we built it

• Mobile Engine: Lightweight, ultra-optimized HTML5/JS core packaged natively for Android using Capacitor 8.
• In-App Subscriptions: Powered by @revenuecat/purchases-capacitor connecting to Google Play and the Samsung
Galaxy Store.
• Deterministic Architecture: Strict 3-state finite state machine with injected clock and storage to prevent
clock tampering.
• Reproducible Packaging: Full headless Docker builder (eclipse-temurin:21-jdk-jammy, Android SDK 36, Build-
Tools 36.0.0) generating verified, cryptographically signed APKs (hachi-release.apk).

## Challenges we ran into

Preserving a clean offline experience while enforcing store subscription validity is notoriously difficult.
Typical implementations either lock the user out the moment connectivity drops, or open the gates to piracy.
We resolved this tension by implementing a mathematically strict offline grace period (72h) tracked by
deterministic timestamps.

## Accomplishments that we're proud of

• Real measured execution: 0.130 ms response time on device. • Zero external cloud dependencies for knowledge queries. • Comprehensive test harness with 100% green verification gates prior to any build.

## What we learned

Local-first AI changes the paradigm of developer monetization. With RevenueCat handling entitlement states,
developers can deliver private, ultra-fast applications without having to manage heavy, costly backends.

## What's next for Arthur

Expanding multi-modal capabilities on mobile while keeping every weight and every token strictly on the user's hardware.

Built With

• ai
• android
• capacitor
• docker
• galaxy
• javascript
• local-first
• revenuecat
• samsung