Unfilter cover
all projects

Personal · Open Source · Released

Unfilter

pinned

An Android app analyzer that reveals tech stacks and frameworks of installed apps, fully offline, with usage, process, and consumption insights.

GitHubPlay Store
  • Flutter
  • Dart
  • Riverpod
  • Platform Channels
  • Fastlane
  • GitHub Actions
  • Play Store

Problem

Modern Android apps are built with diverse frameworks: native, Flutter, React Native, Unity, KMP. But there's no simple, offline-first way to inspect the tech stack behind installed apps along with their runtime behavior and resource usage. Existing tools are developer-only, require root, or depend on online services.

Constraints

  • Fully offline, no root access
  • Low overhead on system performance
  • Compatible with mid-range devices
  • Accurate framework detection despite limited OS-level visibility
  • Clean UI usable by developers and non-developers alike
  • Updates must not depend on Play Store availability

Approach

Designed Unfilter as an offline analyzer that inspects installed packages, infers tech stacks via signature-based detection, and surfaces system insights like memory, storage, and running processes through a clean UI. Architecture cleanly separates package discovery, stack detection, system stats, and update management.

DecisionReasonTrade-off
Offline detection via package inspectionNo privacy issues or network dependencyAccuracy bound by known signatures
Launch-time re-sync over persistent listenersAndroid restricts background listenersExplicit revalidation instead of realtime
Gated OTA updates (GitHub-hosted JSON)Bypass Play Store constraints, controlled rolloutsManual version management
Flutter + platform channelsFast UI iteration with deep system accessComplexity at the native boundary

Hard problems

Detecting installs/removals reliably

Android offers no reliable always-on mechanism without background restrictions or elevated permissions. Implemented a sync-on-launch package re-scan plus a manual "Revalidate" fallback, correctness without violating OS constraints or battery budgets.

OOM crashes on 4GB devices

Intensive scans crashed low-end devices. Fixed with image cache dimension capping via ResizeImage, batched processing limited to 2 concurrent tasks matching the native thread pool, and aggressive cache clearing on widget disposal.

Framework detection accuracy

Obfuscation and diverse build configs break naive detection. Built multi-vector detection combining static, binary, and heuristic analysis over a continuously updated signature database, with confidence scoring and graceful fallback.

Architecture

  • Frontend - Flutter 3.10+ · Riverpod · Material 3 with adaptive dark/light theming
  • Native - Platform channels · NIO file traversal · PackageManager integration · UsageStats API
  • Core - DeviceAppsRepository (TTL caching) · detection engine · performance monitor · OTA update service
  • Data - 30+ field DeviceApp entity · repository pattern · compute-isolated JSON caching · stream-based progress

Stats

  • ~25,000+ lines across Flutter, Dart, and native Android
  • 95% detection accuracy across 10+ major frameworks
  • <150MB peak memory, 2-3s scan for 100 apps
  • Shipped on Play Store + GitHub releases

What I learned

Design around Android's constraints, not against them. Offline-first tools demand trust in local heuristics and graceful failure. Update systems are product features, not deployment mechanics. Performance is adoption-critical on mid-range devices.