Changelog
Track the evolution of Eris. with detailed release notes
v1.7.0
Releasedadded
- iOS 26 Liquid Glass design with full backward compatibility
- ThinkingView - Collapsible panel to see AI reasoning in real-time
- Streaming support for thinking content in reasoning models
- Markdown table rendering support
- Clickable links in markdown
- Strikethrough text support in markdown
changed
- Updated MLX and dependencies to latest versions
- Refreshed 2025 AI models registry
- Applied Liquid Glass effect to model selector and scroll button
fixed
- Improved thread card tappable area and preview text
v1.6.2
Releasedadded
- Qwen3 0.6B - Ultra-lightweight multilingual model with thinking capabilities
- Qwen3 1.7B - Balanced multilingual model with advanced reasoning
- Phi-4 Mini - Microsoft's powerful compact model with GPT-4 class performance
- Qwen Coder 0.5B - Ultra-lightweight code assistant
- Qwen Coder 3B - Advanced code assistant with strong capabilities
changed
- Updated AI models registry with 2025 MLX-compatible models
- Set Qwen3 0.6B as the new default model
- Reorganized models with Legacy tag for older versions
- Improved model sorting: non-legacy models shown first
removed
- Llama 3.2 3B (kept 1B version)
- Qwen 2.5 3B (replaced by Qwen3 series)
- Phi 3.5 Mini (replaced by Phi-4 Mini)
v1.6.1
Releasedadded
- iOS 26 beta warning dialog to alert users about potential compatibility issues
changed
- Updated build version number
v1.6.0
Releasedadded
- QuickChatIntent for conversational AI interactions via Siri Shortcuts
- SummarizeTextIntent for text summarization via Siri Shortcuts
- TranslateTextIntent for multi-language translation via Siri Shortcuts
- GenerateTextIntent for creative text generation via Siri Shortcuts
- Siri integration for all shortcuts with voice activation
- Support for different response styles and tones in shortcuts
- Options to save/discard chats in all shortcuts
- Model selection parameter in shortcuts with clear examples
- isPinned property to Thread model for chat organization
- Pin/unpin action in context menu with 3 chat limit
- Visual pin indicator with 45-degree rotation animation
- Haptic feedback and limit alert when exceeding 3 pins
- DocumentationView with comprehensive sections for getting started, privacy, models, chat features, shortcuts, and tips
- Detail views for AI models, chat features, shortcuts, and tips
- Navigation to documentation from Settings
- Specific error types for unsupported model architectures and missing configurations
- Model validation before download attempts
- Detection for 'stablelm' type models which are not supported by MLX
- Explicit keyboard dismissal after sending message
- IntentError enum for proper error handling in shortcuts
- Device requirements and best practices documentation
changed
- Simplify input area layout with cleaner design
- Move model selector to navigation bar as interactive button
- Increase text field and send button sizes for better usability
- Apply subtle gray background colors for both light/dark themes
- Add visual indicators (chevron) to show model selector is clickable
- Remove redundant UI elements for minimalist interface
- Display pinned chats at the top of the list
- Update error messages to be more descriptive and helpful
- Match app's design language with consistent row-based components in documentation
- Change default text type from email to paragraph in GenerateTextIntent
- Improve parameter descriptions with helpful examples in shortcuts
- Optimize performance with shared LLMEvaluator instance
fixed
- CreditsView SafariView (thanks to Pavel Koupa for PR)
- Button content shape on ModelManagementView and SettingsView (thanks to Pavel Koupa for PR)
- Keyboard stays closed when AI finishes responding
- Compilation errors with proper error types and MainActor annotations
- Ambiguous ModelConfiguration references in shortcuts
removed
- Mistral 7B model due to MLX compatibility issues
- CodeLlama 7B model due to MLX compatibility issues
- StableCode 3B model due to MLX compatibility issues
performance
- Add rounded corners to input container background for smoother rendering
- Include smooth spring animations for pin state changes
- Register shortcuts provider in main app for better performance
v1.5.0
Releasedadded
- Device Model Display: Shows specific device model (e.g. 'iPhone 15 Pro') instead of generic 'iPhone'
- Report Bug or Request Feature Section with GitHub Issues link for bug reports
- Discord Community link for getting help and suggesting features
- Developer Mode toggle below Haptic Feedback in Preferences
- NetworkMonitor to detect WiFi vs cellular connections
- Retry logic with exponential backoff for download failures
- Clear error messages explaining MLX framework limitation on cellular
- Device model name mapping for all iPhone and iPad models
- ChipFamily raw values for proper device comparison
changed
- Replace hardcoded version '1.0.0' with dynamic version from bundle
- Reduce MLX cache limit to 20MB during downloads for better stability
- UI to inform users that WiFi is required only for initial download
- Major refactoring: Create AIModels.swift as single source of truth for all model configurations
- Add model categories (General, Reasoning, Code) with proper organization
- Implement device-specific compatibility checking with RAM requirements
- Move download state management to ModelManager for persistence across views
- Improve UI/UX in model management with clearer information display
- Update all views to use centralized AIModelsRegistry
fixed
- State conflict between Developer Mode toggle and About Developer button
- Download progress tracking when navigating between views
- Show/hide Developer section based on toggle state
- Prevent multiple simultaneous downloads with proper state management
removed
- Redundant GitHub and Discord links from Information section
- DEBUG conditional compilation for developer features
- Obsolete ModelConfiguration+Extensions.swift
performance
- Reduced code duplication across the codebase
- Improved maintainability with centralized model management
- Better user experience with more detailed model information
- More reliable download state management across views
v1.4.0
Releasedadded
- Dynamic GPU cache limits based on device capabilities
- Model compatibility system with visual indicators
- Warnings before downloading risky models
- Improved error messages for Metal compilation failures
changed
- Sort models by device compatibility in UI
- Update documentation to reflect actual device requirements
removed
- iPhone 11 base model support (only 4GB RAM causes crashes)
- Note: iPhone 11 Pro/Pro Max (6GB RAM) and SE 2nd gen remain supported
v1.3.0
Releasedadded
- Stop Generation Button to cancel text generation mid-stream
- Thread-safe CancellationToken class for cross-actor cancellation
- Scroll-to-Bottom floating button with smooth animations
- Memory warning handler to clear caches and reduce GPU memory
- Initialize MemoryManager on app startup for automatic memory handling
- Loading spinner when model is being loaded into memory
- Regex caching for syntax highlighting with NSCache
- Haptic feedback when pressing stop button
- Languages: Dart, Lua, Elixir, Clojure, Zig, Nim, Julia
- Enterprise languages: Groovy/Gradle, PowerShell, Visual Basic .NET
- Scientific languages: Fortran, MATLAB, R
- Legacy: Pascal/Delphi
- Web frameworks: Vue.js, React/JSX/TSX, Svelte, Angular
- Markup/Config: HTML, CSS, YAML, JSON, Markdown
- DevOps: Dockerfile, Terraform, GraphQL
- Mobile: Objective-C, Kotlin
- Blockchain: Solidity
changed
- Reordered parser conditions to check lists before bold text
- Added processInlineMarkdown function to handle bold, italic and code formatting within list items
- Stop button color from blue to red for better visibility
- Improve auto-scroll behavior for new messages
- Differentiate between model loading and text generation states
fixed
- Now correctly renders lists with inline markdown like '- **Title**: Content'
- Prevent stop action during model loading phase
performance
- Reduced regex compilation overhead in code blocks
- Optimize code syntax highlighting with cached regex patterns
- Better memory management under pressure
- More responsive UI with optimized scroll handling
v1.0.0
Releasedadded
- On-device AI chat functionality
- Privacy-focused design with no data leaving your device
- Local model execution using MLX framework
- Multiple AI model support
- Dark mode interface
Stay Updated
Join our Discord community to get notified about new releases and share feedback