Release Notes

Smarter extraction with on-device NLP, a refined timeline, manual corrections and event creation, and voice entry.

Extraction and NLP

• Concise action summaries. The dependency parser ONNX model is now bundled into the app, so actions like "played with sera for 90 minutes" show as "Played with Sera" instead of the full sentence
• Better compound food matching. "Cherry tomatoes" and "baked beans" no longer split into separate items
• Meal deduplication. Saying "had lunch" no longer creates both a meal and a duplicate action event
• Smarter extraction timing. Completed sentences extract after 150ms, trailing fragments wait 1.5s. No more mid-sentence extraction flicker

Timeline

• Lane overlap fix. Events that used to stack on top of each other now pack correctly into side-by-side columns
• Short events look short. A 5-minute event no longer takes the same vertical space as a 2-hour one
• Typing velocity fade. The timeline dims slightly while you're mid-sentence and resolves when extraction completes

Corrections

• Manual time editing. Adjust start time and duration directly in the correction panel
• "Use as-is" for custom foods. Keep your original name and save it to your personal dictionary
• Manual event creation. Add events to the timeline that weren't extracted from text

Voice Entry

• Speech recognizer service. Dictate log entries instead of typing. macOS, microphone permission requested on first use

Settings and Data

• Personalization settings view. New dedicated settings panel for customizing your experience
• Delete All My Data. One-click data deletion in settings
• 90-day auto-cleanup. Diagnostic logs automatically cleaned up after 90 days

A major update with a visual redesign, significantly smarter extraction, new insights tools, Apple Health integration, and native subscriptions.

Visual Redesign

• New design system. Clean, sharp aesthetic with flat cards, vertical swim-lane timeline, always-visible entry dock
• New typeface. Switched to IBM Plex Sans for a more precise, readable feel
• Custom icon set. 17 purpose-built geometric line icons replacing generic system icons
• Refined animations. Smoother transitions throughout the timeline and correction panels

Smarter Extraction

• More accurate categorization. Significantly reduced mis-categorized entries across all categories
• Better exercise recognition. Improved detection of bodyweight exercises, calisthenics, compound movements
• Expanded food vocabulary. Added dozens of common foods that were previously missed
• Meal grouping. Solo food items can now be grouped into meals
• People and places. tinh now picks up mentions of people and places from your entries
• Faster processing. Caching improvements for snappier extraction as you type

Time Detection

• Much more accurate time parsing. Correctly interprets written-out times, time ranges, 24-hour notation, and ambiguous AM/PM situations
• Smarter AM/PM inference. Uses context from your day structure to resolve ambiguous times

Insights and Analytics

• New Insights tab. Explore your data with heatmaps, trends, and pattern analysis
• Weather correlations. See how weather conditions relate to your habits
• Apple Health integration. Pull in data from Apple Health for richer correlations

Corrections and Editing

• Inline corrections panel. Edit extractions directly from the timeline
• "Add to meal" action. Quickly group standalone food items into a meal
• Delete entries. Remove entries from the log book

Subscriptions

• Native subscriptions. Replaced the old license system with native App Store subscriptions via StoreKit 2

Security

• App lock. Optional app lock to keep your health data private

A complete rebuild from the ground up. tinh v0.2 is a native Swift application with real-time data extraction.

Real-Time Data Extraction

• Instant feedback. See categories appear in real-time as you write, under a second per sentence
• 20 seconds per entry. Full entry processing now takes ~20 seconds, down from 5-8 minutes
• No LLM required. ML models run entirely on-device without requiring a large language model

Full Swift Rebuild

• Native performance. Faster startup, lower memory usage, smoother animations
• Better macOS integration. Native menus, keyboard shortcuts, and system features

Redesigned Corrections UI

• Inline corrections. Fix categorizations directly where they appear
• Smarter learning. Corrections propagate more effectively to future entries

Expanded Analytics

• New temporal analysis. Understand when and how often things happen with improved timing visualizations
• 128 custom variables. Supports up to 128 input variables and 128 outcome variables

Visual overhaul, smarter correlation analysis, and rebuilt submission experience.

Dashboard Redesign

• New card layout. Reorganized with Inputs/Outputs in a cleaner two-column layout
• Improved navigation. Updated sidebar with custom brand icons
• Full dark mode. Proper color scheme across all screens

Enhanced Correlation Analysis

• Presence mode. Ask "Do I feel better on days when I do this at all?" vs. amount-based analysis
• Improved food correlations. More reliable results with increased sample requirements
• Exercise integration. Exercise correlations now available with duration/presence toggles

Initial beta release with core log book, correlation analysis, nutrition tracking, and privacy-first architecture.

The original tinh prototype was in development throughout 2024 before being rebuilt from scratch as a native Swift application. The core ideas validated during this period (natural language input, on-device categorization, and correlation discovery) were carried forward into the current architecture.