ZiG v0.2.0 — Smarter, Faster, Fully Redesigned

🚀 ZiG v0.2.0 — Smarter, Faster, Fully Redesigned

The second public release of ZiG (Zen i Guess).

Since v0.1, ZiG has received a complete UI redesign, a significantly smarter on-device decision pipeline, and numerous UX improvements—while staying true to the principles it was built on:

• 🔒 100% on-device processing
• 🚫 No cloud
• 🚫 No telemetry
• 🚫 No accounts
• 🌐 No internet connection required

---

✨ What's New

🎨 Complete UI Redesign

The interface has been rebuilt from the ground up using Jetpack Compose and Material 3.

Highlights include:

• Cleaner and more consistent visual design
• Faster notification triage
• Improved navigation throughout the app
• Reduced friction across the entire experience

📥 Notification Review

Notification Review has been redesigned into a fast, one-tap inbox.

• ✅ Approve or block notifications instantly
• ↩️ Undo is always available
• ⚡ Faster workflow with fewer interactions

---

🧠 Smarter & More Efficient Decision Pipeline

⚡ Deterministic Exact-Match Cache (New)

Before running any AI inference, ZiG now checks whether a notification has already been seen.

If the newest notification message exactly matches one you've previously approved or blocked:

• No embedding generation
• No TensorFlow Lite inference
• No KNN similarity search

Instead, your previous decision is replayed instantly.

The cache is:

• Case-insensitive
• Database index-backed
• Based on the newest message in a notification, allowing repeated messages in long conversations to be recognized correctly

This dramatically reduces latency for recurring notifications while also lowering CPU usage.

---

🧩 Improved Classification

The on-device ensemble has been made significantly more robust.

Changes include:

• Stronger consensus requirements before decisions are made
• Better interaction between the base classifier and Personal Memory
• Reduced influence from sparse or ambiguous user history

This results in more stable predictions while still allowing the model to gradually adapt to your personal preferences.

---

⚠️ Known Limitations

ZiG is still an early preview release.

Areas actively being improved include:

• ML classification accuracy
• Personal Memory adaptation
• Additional rule capabilities
• UI polish
• Performance optimizations

---

💬 Feedback

Bug reports, feature requests, and contributions are always welcome.

Please use GitHub Issues for feedback and bug reports.

---

❤️ Thank You

Thank you for using ZiG.

Your notifications. Your AI. Your phone.

ZiG-v0.1-Release

ZiG v0.1.0 — Initial Public Release

Welcome to the first public release of ZiG (Zen i Guess).

ZiG is a privacy-first Android notification filter that intelligently decides which notifications deserve your attention—entirely on your device. No cloud services, no telemetry, no accounts, and no internet access.

What's Included
Privacy by Design
All notification processing happens locally.
No Internet permission.
No analytics or telemetry.
No notification data ever leaves your phone.
No third-party cloud AI services.
Smart Notification Pipeline

Every notification passes through a layered decision pipeline designed for speed and battery efficiency:

Managed Apps filtering
Contact whitelist
Custom keyword rules
On-device machine learning

Deterministic checks run first, ensuring the AI is only invoked when necessary.

On-Device Machine Learning

When rules aren't enough, ZiG uses an entirely local ML ensemble to classify notifications.

The classifier combines:

A custom TensorFlow Lite notification classifier
Personal Memory powered by on-device embeddings
Similarity-based learning from your previous decisions

Nothing is uploaded. Everything stays on your device.

Personal Memory

ZiG gradually adapts to how you handle notifications.

Manual overrides become local training examples that improve future predictions while never leaving your phone.

Managed Apps

Choose exactly which applications ZiG manages.

Apps you don't explicitly enable remain completely untouched.

Custom Rules

Create deterministic keyword rules that bypass AI entirely.

Examples:

OTP
PNR
Boarding Pass
cab, arriving

Rules support multi-keyword matching for precise filtering.

Daily Summary

Receive a once-per-day summary of filtered notifications through a local notification.

No cloud scheduling.

No external services.

Native Rust Engine

Performance-critical components are implemented in Rust, including:

Contact lookup
Managed app filtering
Keyword matching

This provides extremely fast decision making while keeping memory usage low.

Built With
Kotlin
Jetpack Compose
Material 3
Rust (JNI)
TensorFlow Lite
MediaPipe Text Embedder
Room
Hilt
MVVM Architecture
Known Limitations

This is the first public release and should be considered an early preview.

Areas that will continue to improve include:

ML classification accuracy
Personal Memory adaptation
UI polish
Additional rule capabilities
Performance optimizations
Feedback

Bug reports, feature requests, and contributions are welcome.

GitHub Issues are the preferred place for reporting problems or suggesting improvements.

Thank you for trying ZiG.

Your notifications. Your AI. Your phone.