Skip to content

CHANGELOG.md

Latest commit

 

History

History
675 lines (538 loc) · 36.3 KB

CHANGELOG.md

File metadata and controls

675 lines (538 loc) · 36.3 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Added

  • Cancel Processing Button: Added ability to cancel ongoing transcription processing
  • Dark Mode Icon Inversion: Monochrome provider icons now automatically invert in dark mode for better visibility
  • Dynamic Window Resizing: Window automatically resizes based on command menu and toast visibility

Changed

  • Design System Overhaul: Complete refactor of styling to use design tokens throughout the codebase
    • Button component now uses text-foreground, bg-muted, border-border instead of hardcoded hex values
    • Removed hardcoded classes and inline styles across components
    • Improved button and badge consistency
  • Settings UI Redesign: Overhauled all settings pages with unified panel system, redesigned sidebar, and extracted permissions section
  • Dark Mode Polish: Premium button styling, glass morphism toasts, and streamlined visuals
  • Onboarding Improvements:
    • Streamlined to 4-step flow
    • Extended to support multiple local transcription providers (Whisper and NVIDIA Parakeet)
    • Improved contrast and design system usage

Fixed

  • Light Mode UI Visibility: Fixed multiple UI elements that were invisible or hard to see in light mode:
    • Settings gear icon in permission cards now uses text-foreground
    • Troubleshoot button uses proper foreground color
    • Reset button in developer settings now correctly shows destructive color
    • Settings and Help icons in the toolbar are now properly visible
    • Check for Updates button now renders correctly in light mode
  • Provider Tab Flashing: Resolved TranscriptionModelPicker tab flashing by extracting ModeToggle component and syncing internal state with props
  • Local Reasoning Model Persistence: Fixed local reasoning model selection not persisting correctly
  • Parakeet Model Status: Added dedicated IPC channel for Parakeet model status checks
  • Groq Qwen3 Models: Removed thinking tokens from Qwen3 models on Groq provider

[1.3.3] - 2026-01-28

Added

  • ONNX Warm-up Inference: Parakeet server now runs warm-up inference on start to eliminate first-request latency from JIT compilation
  • Startup Preferences Sync: Renderer startup preferences are now synced to .env for server pre-warming on restart

Changed

  • macOS Tray Behavior: Hide to tray on macOS for consistent cross-platform behavior

Fixed

  • macOS Launch Crash: Added disable-library-validation entitlement to resolve macOS launch crash (#120)
  • Reasoning Model Default: Fixed useReasoningModel not correctly defaulting to enabled by persisting useLocalStorage defaults and aligning direct reads
  • Windows Non-ASCII Usernames: Resolved whisper-server crash on Windows with non-ASCII usernames by pre-converting audio to WAV and routing temp files through ASCII-safe directory
  • Windows Paths with Spaces: Fixed temp directory fallback to also detect paths with spaces on Windows

[1.3.2] - 2026-01-27

Changed

  • Linux Paste Tools: Prefer xdotool over ydotool for better compatibility

Fixed

  • Windows Zip Extraction: Use tar instead of PowerShell Expand-Archive for zip extraction on Windows to avoid issues with special characters

[1.3.1] - 2026-01-27

Changed

  • Download System Refactor: Consolidated model download logic into shared utilities with resume support, retry logic, abort signals, and improved installing state UI
  • Throttled Progress Display: Whisper model download progress updates are now throttled for smoother UI

[1.3.0] - 2026-01-26

Added

  • NVIDIA Parakeet Support: Fast local transcription via sherpa-onnx runtime with INT8 quantized models
    • parakeet-tdt-0.6b-v3: Multilingual (25 languages), ~680MB
  • Windows Push-to-Talk: Native Windows key listener with low-level keyboard hook for true push-to-talk functionality
    • Supports compound hotkeys like Ctrl+Shift+F11 or CommandOrControl+Space
    • Prebuilt binary automatically downloaded from GitHub releases
    • Fallback to tap mode if binary unavailable
  • Custom Dictionary: Improve transcription accuracy for specific words, names, and technical terms
    • Add custom words through Settings → Custom Dictionary
    • Words are passed as hints to Whisper for better recognition
    • Works with both local and cloud transcription
  • GitHub Actions Workflow: Automated CI workflow to build and release Windows key listener binary
  • Shared Download Utilities: New scripts/lib/download-utils.js module with reusable download, extraction, and GitHub release fetching functions

Changed

  • Download Scripts Refactored: All download scripts now use shared utilities for consistency
  • GitHub API Authentication: Download scripts support GITHUB_TOKEN to avoid API rate limits in CI
  • Debug Logging Cleanup: Extracted common window loading code and cleaned up debug logging

Fixed

  • GNOME Wayland Hotkey Improvements: Improved hotkey handling on GNOME Wayland
  • Hotkey Persistence: Fixed hotkey selection not persisting correctly
  • Custom Endpoint API Keys: Fixed custom endpoint API keys not persisting to .env file
  • Custom Endpoint State: Fixed custom endpoint using shared state instead of its own
  • Linux Stale Hotkey Registrations: Clear stale hotkey registrations on startup on Linux
  • Wayland XWayland Paste: Try xdotool on Wayland when XWayland is available
  • llama-server Libraries: Bundle llama-server shared libraries and search from extract root for varying archive structures
  • STT/Reasoning Debug Logging: Added missing debug logging for STT and reasoning pipelines

[1.2.16] - 2026-01-24

Fixed

  • App Startup Hang: Fixed app initialization timing issues with Electron 36+
  • Manager Initialization: Deferred manager initialization until after app.whenReady() to prevent hangs
  • Debug Logger Initialization: Deferred debugLogger file initialization until app.whenReady()
  • Config Bundling: Fixed missing config files in production builds
  • whisper.cpp Binary Version: Updated whisper.cpp release names and bumped binary version

[1.2.15] - 2026-01-22

Added

  • ydotool Fallback for Linux: Added ydotool as additional fallback option for clipboard paste operations on Linux systems

Changed

  • Unified Prompt System: Refactored to single intelligent prompt system for improved consistency and maintainability
  • whisper.cpp Remote: Refactored remote whisper.cpp integration for better reliability

[1.2.14] - 2026-01-22

Added

  • Troubleshooting Mode: New debug logging section in settings with toggle for detailed diagnostic logs, log file path display, and direct folder access for easier support
  • Custom Transcription Endpoint: Support for custom OpenAI-compatible transcription endpoints with configurable base URLs
  • Enhanced Clipboard Debugging: Detailed clipboard operation logging for diagnosing paste issues across platforms

Changed

  • API Key Management: Consolidated and refactored API key persistence with improved .env file handling and recovery mechanisms
  • Local Network Detection: Refactored URL detection into reusable utility for better code organization
  • Electron Builder: Updated to latest version for improved build performance

Fixed

  • Windows/Linux Taskbar: Prevented dual taskbar entries on Windows and Linux by properly configuring window behavior
  • Single Instance Lock: Enforced single instance lock with cleaner window state checks
  • Model Provider Consistency: Removed redundant fallbacks and ensured consistent use of getModelProvider()
  • Cross-env Support: Fixed Windows compatibility in pack script using cross-env
  • Linux X11 Paste: Improved paste reliability by capturing target window ID upfront with windowactivate --sync, added xdotool type fallback for terminals
  • Tray Minimize: Fixed minimize to tray functionality

[1.2.12] - 2026-01-20

Added

  • LLM Download Cancellation: Added ability to cancel in-progress local LLM model downloads with throttled progress updates to prevent UI flashing

Changed

  • Gemini Model Updates: Updated Gemini models to latest versions
  • Linux Wayland Improvements: Improved Wayland paste detection with GNOME-specific handling and XWayland fallback support
  • whisper.cpp CUDA Support: Updated whisper.cpp download script to include CUDA-enabled binaries

Fixed

  • Windows Paste Delay: Adjusted paste delay timing on Windows for more reliable text insertion
  • Blank Audio Prevention: Fixed issue where blank/silent audio recordings would paste empty text
  • Newline Handling: Fixed newline formatting issues in transcribed text

[1.2.11] - 2026-01-18

Fixed

  • ASAR Path Resolution: Fixed path resolution issues for bundled resources in packaged builds
  • Update Checker: Fixed auto-update checker initialization
  • Build Includes: Ensured services and models are properly included in production builds
  • OS Module Import: Fixed OS module import ordering

[1.2.10] - 2026-01-17

Fixed

  • Streaming Backpressure: Fixed proper streaming backpressure handling in audio processing
  • Quit and Install: Fixed update installation on app quit

[1.2.9] - 2026-01-17

Fixed

  • Path Resolution: Improved path resolution for better cross-platform compatibility

[1.2.8] - 2026-01-16

Added

  • Microphone Input Selection: Choose your preferred microphone input device in settings, with built-in mic preference to prevent Bluetooth audio interruptions
  • Push to Talk Mode: New recording mode option alongside the existing toggle mode
  • Hotkey Listening Mode: Prevents conflicts when capturing new hotkeys by temporarily disabling the global hotkey
  • Hotkey Fallback System: Automatic fallback with user notifications when preferred hotkey is unavailable
  • Cross-Platform Accessibility Settings: Quick access to system accessibility settings on macOS

Changed

  • Streamlined Onboarding: Removed redundant "How it Works" section, success dialogs, and manual save buttons for a smoother setup experience
  • Improved Select Styling: Enhanced dropdown select component appearance

Fixed

  • FFmpeg Availability Types: Corrected type definitions and optimized whisper-cpp download process
  • Whisper Models Path: Fixed model storage path resolution
  • Better Path Resolution: Improved error handling for file paths
  • Open Mic Settings: Fixed system settings link for microphone configuration

[1.2.7] - 2026-01-13

Added

  • Whisper Server HTTP Mode: Added persistent whisper-server for faster repeated transcriptions with automatic CLI fallback
  • Pipeline Timing Instrumentation: Added detailed timing logs for each stage of the transcription pipeline
  • Whisper Server Pre-warming: Server pre-warms on startup for faster first transcription

Changed

  • Windows Clipboard: Reduced clipboard delays for faster text pasting on Windows

Fixed

  • Windows Update Install: Simplified Windows update installation by using silent mode and removing redundant before-quit handling
  • Mac Build Workflows: Fixed CI/CD to run separate workflows for Mac builds
  • Mac DMG Build Race Condition: Fixed release workflow DMG build failure caused by concurrent arm64/x64 builds mounting same volume
  • Windows Download Script: Fixed PowerShell Expand-Archive failure with bracket characters in directory names

[1.2.6] - 2026-01-13

Changed

  • Settings Layout: Moved settings navigation to left side on Windows and Linux for improved consistency

Fixed

  • Linux Whisper Detection: Fixed issue where Python-based Whisper could be used instead of whisper.cpp on Linux systems

[1.2.5] - 2026-01-13

Added

  • Model Validation: Added validation when deleting or loading Whisper models to ensure model integrity
  • Download Cancellation: Added ability to cancel in-progress model downloads in whisper pickers
  • Windows Paste Performance: Added nircmd for faster text pasting on Windows

Fixed

  • EventEmitter Memory Leak: Fixed memory leak caused by duplicate listener registration in useUpdater hook across ControlPanel and SettingsPage components
  • FFmpeg Path Resolution: Fixed FFmpeg path resolution in unpacked ASAR for local whisper.cpp transcription

Changed

  • UI Cleanup: Removed redundant UI elements for a cleaner interface

[1.2.4] - 2026-01-13

Changed

  • whisper.cpp Packaging: Moved whisper.cpp binaries from ASAR to extraResources for improved reliability and faster startup

Fixed

  • Package Lock Sync: Fixed package-lock.json synchronization with package.json dependencies

[1.2.3] - 2026-01-13

Added

  • Extended Hotkey Support: Added numpad keys, media keys, and additional special keys (Pause, ScrollLock, PrintScreen, NumLock) for hotkey selection
  • Improved Hotkey Error Messages: Registration failures now include helpful suggestions for alternative hotkeys

Changed

  • Linux Paste Tools: Only show paste tools installation prompt on Linux when tools are not available

Fixed

  • Hotkey Debugging: Added comprehensive debug logging to hotkey manager for troubleshooting registration issues

[1.2.2] - 2026-01-13

Fixed

  • React Version Mismatch: Fixed blank screen caused by incompatible React and React-DOM versions in package-lock.json

[1.2.1] - 2026-01-13

Fixed

  • Blank Screen on Upgrade: Fixed white screen issue for users upgrading from older versions with different onboarding step counts. The onboarding step index is now properly clamped to valid range.

[1.2.0] - 2026-01-13

Added

  • Delete All Whisper Models: New option to delete all downloaded Whisper models at once
  • Model Deletion Confirmation: Added confirmation dialog when deleting models in settings

Changed

  • Migrated to whisper.cpp: Replaced Python-based Whisper with native whisper.cpp for faster, more reliable transcription
    • No longer requires Python installation
    • WebM-to-WAV audio conversion built-in
    • Significantly improved startup and transcription speed
  • Streamlined Onboarding: Simplified setup flow with fewer steps now that Python is not required
  • Download Cancellation: Added ability to cancel in-progress model downloads
  • CI/CD Updates: Updated build and release workflows

Fixed

  • IPC Handler: Fixed broken IPC handler for model operations
  • Logging: Standardized logging across the application
  • React Hook Dependencies: Improved React hook dependency arrays for better performance
  • Button Styling: Fixed button styling consistency across the application

Removed

  • Python Dependency: Removed Python requirement and all related installation code
  • whisper_bridge.py: Removed Python-based Whisper bridge in favor of native whisper.cpp

[1.1.2] - 2026-01-12

Added

  • Linux Package Dependencies: Recommended xdotool, wtype, and python3 packages for Linux users

Fixed

  • Python Installation Race Condition: Fixed race condition in Python installation check that could cause installation to fail or hang

[1.1.1] - 2026-01-12

Added

  • Cross-Platform Paste Tools Detection: Onboarding now detects and guides users through installing paste tools on Linux and Windows with auto-grant accessibility

Changed

  • Qwen Model Compatibility: Disabled thinking mode for Qwen models on Groq to prevent compatibility issues
  • Model Registry Refactor: disableThinking flag now uses the centralized model registry
  • Consolidated ColorScheme Types: Removed redundant default exports and cleaned up inline font styles
  • Provider Icons: Use static imports for provider icons to fix Vite bundling issues

Fixed

  • Recording Cancellation: Restored cancel recording functionality that was accidentally removed
  • Model Downloads: Implemented atomic downloads with temp file pattern and robust cleanup handling for cross-platform reliability
  • Incomplete Download Prevention: Model file size validation now prevents incomplete downloads from showing as complete
  • Windows PowerShell Performance: Optimized paste startup time on Windows

[1.1.0] - 2026-01-10

Added

  • Compound Hotkey Support: Use multi-key combinations like Cmd+Shift+K or Ctrl+Alt+D for dictation
  • Groq API Integration: Ultra-fast AI inference with Groq's cloud API
  • Auto-Update UI: Download progress bars and install button in settings
  • Recording Cancellation: Cancel an in-progress recording without transcribing
  • Release Notes Viewer: Markdown-rendered release notes in settings

Changed

  • Major Hotkey Refactor: Complete rewrite of hotkey selection with improved reliability and validation
  • Consolidated Model Registry: Single source of truth for all AI models (modelRegistryData.json)
  • Unified Model Picker: Reusable component for both transcription and reasoning model selection
  • Improved Latency Logging: Numbered stage logs for recording, transcription, reasoning, and paste timing
  • Reduced Paste Delay: Lowered from 100ms to 50ms for faster text insertion
  • Code Quality: Added ESLint, Prettier for JS/TS, and Ruff for Python

Fixed

  • Windows 11 Compatibility: Fixed PATH separator, cache directories, and process termination
  • Python Virtual Environment: Fixed race condition and added Arch Linux venv support
  • Microphone Detection: Improved onboarding flow for missing inputs with deep-linking to system settings
  • Recording State Alignment: Recording now aligns to MediaRecorder's actual start/stop events
  • Caching Optimizations: Cached accessibility, paste tool, and FFmpeg checks to reduce process spawns
  • Window Titles: Electron window titles now set correctly after page load

[1.0.15] - 2026-01-05

Added

  • Button to fully quit OpenWhispr processes from the application
  • Linux terminal detection with automatic paste key switching (Ctrl+Shift+V for terminals)

Changed

  • Standardized logging on log levels with renderer IPC and .env refresh for consistent debug output

Fixed

  • Use kdotool for Wayland terminal detection, improving clipboard paste reliability
  • Increased delay before restoring clipboard to avoid race conditions during paste operations
  • Persist OpenAI key before onboarding test to prevent key loss during setup
  • Windows Python discovery now correctly handles output parsing
  • Keep FFmpeg debug schema as boolean type
  • Fixed OpenWhispr documentation paths
  • Windows: Resolved issue #16 with WAV validation, registry-based Python detection, and normalized FFmpeg paths

[1.0.13] - 2025-12-24

Added

  • Enhanced Linux support with Wayland compatibility, multiple package formats (AppImage, deb, rpm, Flatpak), and native window controls
  • Auto-detect existing Python during onboarding and gate the installer with a recheck option
  • "Use Existing Python" skip flow to onboarding with confirmation dialog

Changed

  • Reuse audio manager and stabilize dictation toggle callback to fix recording latency
  • Add cleanup functions to IPC listeners to prevent memory leaks
  • Make Flatpak opt-in for local builds only

Fixed

  • Optimized transcription pipeline with caching, batched reads, and non-blocking operations for improved performance
  • Reference error in settings page
  • Removed redundant audio listener causing unnecessary processing
  • Added IPC listener cleanup to prevent memory leaks
  • Performance improvements: removed duplicate useEffect, fixed blur causing re-renders

CI/CD

  • Add caching for Electron and Flatpak downloads
  • Add Flatpak runtime installation to workflow
  • Add Linux packaging dependencies to GitHub Actions workflow

[1.0.12] - 2025-11-13

Added

  • Added scripts/complete-uninstall.sh plus a new TROUBLESHOOTING guide so you can collect arch diagnostics, clean caches, and reset permissions before reinstalling stubborn builds.
  • Control Panel history now auto-refreshes through a shared store and IPC events, so new, deleted, or cleared transcripts sync instantly without a manual refresh.
  • Distribution artifacts now include both Apple Silicon and Intel macOS DMG/ZIP outputs, and the README documents Debian/Ubuntu packaging along with optional xdotool support.

Changed

  • The onboarding flow now validates dictation hotkeys before letting you continue, remembers whether cloud auth was skipped, and only persists sanitized API keys once supplied.
  • History entries normalize timestamps and no longer run the removed legacy text cleanup helper, so the UI shows the exact Whisper output that was saved.

Fixed

  • Local Whisper now finds Python on Windows more reliably by scanning typical install paths, honoring OPENWHISPR_PYTHON, and surfacing actionable ENOENT guidance.
  • Whisper installs automatically retry pip operations that hit PEP‑668, TOML, or permission errors, sanitizing the output and falling back to --user + legacy resolver when needed.

[1.0.11] - 2025-10-13

Added

  • Settings, onboarding, and the AI model selector now accept OpenAI-compatible custom base URLs for both transcription and reasoning providers, complete with validation and reset helpers.
  • Windows now gets full tray behavior: closing the control panel hides it to the tray, left-click reopens it, and the UI adds a native close button.

Changed

  • ReasoningService sends both input and messages payloads and automatically falls back between /responses and /chat/completions so older OpenAI-compatible endpoints keep working.

Fixed

  • Successful endpoint detection is cached per base URL, so the app remembers whether to call /responses or /chat/completions instead of retrying the wrong path forever.
  • Custom endpoint fields now enforce HTTPS (with localhost as the lone exception) across the UI and services, preventing API keys from ever leaving over plain HTTP.

[1.0.10] - 2025-10-07

Added

  • Added a compile:globe build step that emits a macOS Globe listener binary into resources/bin before every dev, pack, or dist command so the hotkey ships with all builds.

Fixed

  • Globe key failures now raise a macOS dialog, verify the bundled binary is executable, and kill/restart the listener cleanly so the shortcut survives packaging.

[1.0.9] - 2025-10-07

Changed

  • Simplified the release workflow by removing the bespoke GitHub release job and letting electron-builder upload draft releases directly.

[1.0.8] - 2025-10-03

Fixed

  • Globe/Fn hotkey reliability improved by showing the dictation panel before toggling, making focus optional, and surfacing listener spawn errors instead of failing silently.

[1.0.7] - 2025-10-03

Added

  • Settings update controls now show download progress bars, install countdowns, and clearer messaging while fetching or installing new builds.

Changed

  • Auto-update internals now track listeners, cache the last release metadata, and keep auto-download/auto-install disabled until the user explicitly triggers an update, eliminating the previous memory leaks.

Fixed

  • Install & Restart now emits before-quit, enables autoInstallOnAppQuit, logs progress, and calls quitAndInstall(false, true) so updates actually apply when quitting or pressing the button.

[1.0.6] - 2025-09-11

Added

  • Dictation Panel Command Menu: Clicking the floating panel reveals quick actions, including a one-click "Hide this for now" option.
  • macOS Globe Key Support: Added a lightweight Swift listener so the Globe/Fn key can toggle dictation across the system.
  • Globe Key Selection UI: Settings and onboarding keyboards now include a dedicated Globe key option.
  • Hotkey Validation: Settings and onboarding now verify shortcut registration immediately, alerting users when a key can’t be bound.
  • Model Cache Cleanup: Added an in-app command (and installer/uninstaller hooks) to delete all cached Whisper models.
  • Tray Controls: macOS tray menu gained quick actions to show or hide the dictation panel.

Changed

  • Dictation Overlay Placement: Window now anchors to the active workspace's bottom-right corner with a safety margin, preventing it from sliding off-screen on multi-monitor setups.
  • Dictation Overlay Canvas: Enlarged the floating window so tooltips, menus, and error states render without being clipped while keeping click-through behaviour outside interactive elements.
  • Keyboard UX: Virtual keyboard hides macOS-exclusive keys on Windows/Linux and standardises hotkey labels.

Fixed

  • macOS Window Lifecycle: Ensured the dictation panel keeps the app visible in Dock and Command-Tab while retaining floating behaviour across spaces.
  • Control Panel Stability: Reworked close/minimize handling so the panel stays interactive when switching apps and reopens cleanly without spawning duplicate windows.
  • Always-On-Top Enforcement: Centralised the logic that reapplies floating window levels, eliminating redundant timers and focus quirks.
  • Menu Labelling: macOS application menu items now display the correct OpenWhispr casing instead of "open-whispr".
  • Non-mac Hotkey Guard: Prevented the mac-only Globe shortcut from being saved on Windows/Linux.

[1.0.5] - 2025-09-10

Fixed

  • Build System: Fixed native module signing conflicts on macOS
    • Added npmRebuild: true to force rebuild of native modules during packaging
    • Added buildDependenciesFromSource: true to compile native dependencies from source
    • Added better-sqlite3 to asarUnpack array to properly unpack SQLite3 native module
    • Resolves "different Team IDs" error when launching notarized macOS apps
  • CI/CD Pipeline: Fixed automated release workflow issues
    • Removed automatic version update step from release workflow (version should be set before tagging)
    • Added contents: write permission to allow workflow to create GitHub releases
    • Fixes "Resource not accessible by integration" error during releases

Technical Details

  • This is a maintenance release focusing on build reliability and deployment infrastructure
  • No feature changes or user-facing functionality updates
  • All changes related to packaging, signing, and automated release processes

[1.0.4] - 2025-09-09

Added

  • Multi-Provider AI Support: Integrated three major AI providers for text processing
    • OpenAI: Complete model suite including:
      • GPT-5 Series (Nano/Mini/Full) - Latest generation with deep reasoning
      • GPT-4.1 Series (Nano/Mini/Full) - Enhanced coding, 1M token context, June 2024 knowledge
      • o-series (o3/o3-pro/o4-mini) - Advanced reasoning models with extended thinking time
      • GPT-4o/4o-mini - Multimodal models with vision support
    • Anthropic: Claude Opus 4.1, Sonnet 4, and 3.5 variants for frontier intelligence
    • Google: Gemini 2.5 Pro/Flash/Flash-Lite and 2.0 Flash for advanced processing
  • OpenAI Responses API Integration: Migrated from Chat Completions to the new Responses API
    • Simplified request format with input array instead of messages
    • New response parsing for output items with typed content
    • Automatic handling of model-specific requirements
    • Better support for GPT-5 and o-series reasoning models
  • Enhanced Reasoning Service: Complete TypeScript rewrite with provider abstraction
    • Automatic provider detection based on selected model
    • Secure API key caching with TTL
    • Unified retry strategies across all providers
    • Provider-specific token optimization (up to 8192 for Gemini)
  • Comprehensive Debug Logging: Enhanced reasoning pipeline with stage-by-stage logging
    • Provider selection and routing logs
    • API key retrieval and validation logs
    • Request/response details for all providers
    • Error tracking with full stack traces
  • Improved Settings UI: Comprehensive API key management for all providers
    • Color-coded provider sections (OpenAI=green, Anthropic=purple, Gemini=blue)
    • Inline API key validation and secure storage
    • Provider-specific model selection with descriptions

Changed

  • Default AI Model: Updated from GPT-3.5 Turbo to GPT-4o Mini for cost-efficient multimodal support
  • Model Updates: Refreshed all AI models to their latest 2025 versions
    • OpenAI: Added GPT-5 family (released August 2025), migrated to Responses API
    • Anthropic: Updated to Claude Opus 4.1 and Sonnet 4, fixed model naming
    • Gemini: Added latest 2.5 series models, increased token limits
  • ReasoningService: Migrated from JavaScript to TypeScript for better type safety
  • API Endpoint Updates:
    • OpenAI: Migrated from /v1/chat/completions to /v1/responses
    • Request format simplified for better performance
    • Response parsing updated for new output structure
  • Model Configuration Improvements:
    • Fixed Anthropic model names (using hyphens instead of dots)
    • Increased Gemini 2.5 Pro token limits (2000 minimum)
    • Removed temperature parameter for GPT-5 and o-series models
  • Documentation: Updated CLAUDE.md, README.md with comprehensive provider information

Fixed

  • API Key Persistence: All provider keys now properly save to .env file
    • Added saveAllKeysToEnvFile() method for consistent persistence
    • Keys reload automatically on app restart
    • Fixed Gemini and Anthropic key storage issues
  • CORS Issues: Anthropic API calls now route through IPC handler
    • Avoids browser CORS restrictions in renderer process
    • Proper error handling in main process
  • Empty Response Handling: Fixed "No text transcribed" error when AI returns empty
    • Falls back to original text when API returns nothing
    • Properly handles edge cases in response parsing
  • Parameter Compatibility: Fixed OpenAI API parameter errors
    • GPT-5 models use simplified parameters (no max_tokens)
    • o-series models configured without temperature
    • Older models retain full parameter support

Technical Improvements

  • Added Gemini API integration with proper authentication flow
  • Implemented SecureCache utility for API key management
  • Enhanced IPC handlers for multi-provider support
  • Updated environment manager with Gemini key storage
  • Improved error handling with provider-specific messages
  • Added comprehensive retry logic with exponential backoff
  • Enhanced error messages with detailed logging
  • Better fallback strategies for API failures
  • Improved response validation and parsing
  • Centralized API configuration in constants file
  • Unified debugging system across all providers

[1.0.3] - 2024-12-20

Added

  • Local AI Models: Integration with community models for complete privacy
    • Support for Llama, Mistral, and other open-source models
    • Local model management UI with download progress
    • Automatic model validation and testing
  • Enhanced Security: Improved API key storage and management
    • System keychain integration where available
    • Encrypted localStorage fallback
    • Automatic key rotation support

Fixed

  • Resolved issues with Whisper model downloads on slow connections
  • Fixed clipboard pasting reliability on Windows 11
  • Improved error messages for better debugging
  • Fixed memory leaks in long-running sessions

Changed

  • Optimized audio processing pipeline for 30% faster transcription
  • Reduced app bundle size by 15MB through dependency optimization
  • Improved startup time by lazy-loading heavy components

[1.0.2] - 2024-12-19

Added

  • Automatic Python Installation: The app now detects and offers to install Python automatically
    • macOS: Uses Homebrew if available, falls back to official installer
    • Windows: Downloads and installs official Python with proper PATH configuration
    • Linux: Uses system package manager (apt, yum, or pacman)
  • Enhanced Developer Experience:
    • Added MIT LICENSE file
    • Improved documentation for personal vs distribution builds
    • Added FAQ section to README
    • Added security information section
    • Clearer prerequisites and setup instructions
    • Added comprehensive CLAUDE.md technical reference
  • Dock Icon Support: App now appears in the dock with activity indicator
    • Changed LSUIElement from true to false in electron-builder.json
    • App shows in dock on macOS with the standard dot indicator when running

Changed

  • Updated supported language count from 90+ to 58 (actual count in codebase)
  • Improved README structure for better open source experience

[1.0.1] - 2024-XX-XX

Added

  • Agent Naming System: Personalize your AI assistant with a custom name for more natural interactions
    • Name your agent during onboarding (step 6 of 8)
    • Address your agent directly: "Hey [AgentName], make this more professional"
    • Update agent name anytime through settings
    • Smart AI processing distinguishes between commands and regular dictation
    • Clean output automatically removes agent name references
  • Draggable Interface: Click and drag the dictation panel to any position on screen
  • Dynamic Hotkey Display: Tooltip shows your actual hotkey setting instead of generic text
  • Flexible Hotkey System: Fixed hardcoded hotkey limitation - now fully respects user settings

Changed

  • [BREAKING] Removed click-to-record functionality to prevent conflicts with dragging
  • UI Behavior: Recording is now exclusively controlled via hotkey (no accidental triggering)
  • Tooltip Text: Shows "Press {your-hotkey} to speak" with actual configured hotkey
  • Cursor Styles: Changed to grab/grabbing cursors to indicate draggable interface

Fixed

  • Hotkey Bug: Fixed issue where hotkey setting was stored but not actually used by global shortcut
  • Documentation: Updated all docs to reflect current UI behavior and hotkey system
  • User Experience: Eliminated confusion between drag and click actions

Technical Details

  • Agent Naming Implementation:
    • Added centralized agent name utility (src/utils/agentName.ts)
    • Enhanced onboarding flow with agent naming step
    • Updated ReasoningService with context-aware AI processing
    • Added agent name settings section with comprehensive UI
    • Implemented smart prompt generation for agent-addressed vs regular text
  • Added IPC handlers for dynamic hotkey updates (update-hotkey)
  • Implemented window-level dragging using screen cursor tracking
  • Added real-time hotkey loading from localStorage in main dictation component
  • Updated WindowManager to support runtime hotkey changes
  • Added proper drag state management with smooth 60fps window positioning
  • Code Organization: Extracted functionality into dedicated managers and React hooks:
    • HotkeyManager, DragManager, AudioManager, MenuManager, DevServerManager
    • useAudioRecording, useWindowDrag, useHotkey React hooks
    • WindowConfig utility for centralized window configuration
    • Reduced WindowManager from 465 to 190 lines through composition pattern

[0.1.0] - 2024-XX-XX

Added

  • Initial release of OpenWhispr (formerly OpenWispr)
  • Desktop dictation application using OpenAI Whisper
  • Local and cloud-based speech-to-text transcription
  • Real-time audio recording and processing
  • Automatic text pasting via accessibility features
  • SQLite database for transcription history
  • macOS tray icon integration
  • Global hotkey support (backtick key)
  • Control panel for settings and configuration
  • Local Whisper model management
  • OpenAI API integration
  • Cross-platform support (macOS, Windows, Linux)

Features

  • Speech-to-Text: Convert voice to text using OpenAI Whisper
  • Dual Processing: Choose between local processing (private) or cloud processing (fast)
  • Model Management: Download and manage local Whisper models (tiny, base, small, medium, large)
  • Transcription History: View, copy, and delete past transcriptions
  • Accessibility Integration: Automatic text pasting with proper permission handling
  • API Key Management: Secure storage and management of OpenAI API keys
  • Real-time UI: Live feedback during recording and processing
  • Global Hotkey: Quick access via customizable keyboard shortcut
  • Database Storage: Persistent storage of transcriptions with SQLite
  • Permission Management: Streamlined macOS accessibility permission setup

Technical Stack

  • Frontend: React 19, Vite, TailwindCSS, Shadcn/UI components
  • Backend: Electron 36, Node.js
  • Database: better-sqlite3 for local storage
  • AI Processing: OpenAI Whisper (local and API)
  • Build System: Electron Builder for cross-platform packaging

Security

  • Local-first approach with optional cloud processing
  • Secure API key storage and management
  • Sandboxed renderer processes with context isolation
  • Proper clipboard and accessibility permission handling