All notable changes to AI Text Tools will be documented in this file.

• Reasoning model think tags appearing in output - Local models (DeepSeek R1, Qwen QwQ, etc.) that use <think>...</think> chain-of-thought tags now have these stripped automatically from all provider responses

• Beta tester access broken after DPAPI security update - Beta testers could not use the app because BetaManager was initialised after the Orchestrator, so beta access was never detected at startup
  • Moved BetaManager.Init() before Orchestrator.Initialize() so the provider factory can detect beta tokens
  • Added beta access check to AutoConnectProvider() so beta testers get their provider refreshed on startup
• Beta re-registration now resumes existing session - Activating with an already-registered email now behaves identically to a fresh registration: shows the same activation message, restores remaining credits, and immediately switches to the beta provider
• Beta registration handles server response variations - Token extraction now works regardless of server success flag; handles null uses_remaining by querying status endpoint
• Fixed crash when UsesRemaining is empty - Integer() call in GetUsesRemaining() now handles empty strings gracefully
• Beta registration popup closes before confirmation - Popup now destroys before showing success message for cleaner UX

• Ollama provider rewritten for reliability - Now uses curl.exe (built into Windows 10 1803+) for long-running LLM requests
  • Eliminates 0x8000000A "data not yet available" errors with WinHTTP async mode
  • Falls back to WinHTTP sync mode if curl is unavailable
  • Supports 5-minute timeout for slow model inference
  • Uses UTF-8-RAW encoding to avoid BOM issues with JSON
• HTTP client switched to synchronous mode - More reliable than async for AI API calls; timeouts now set before Open() for proper effect

• DPAPI encryption for API keys - Credentials are now encrypted using Windows Data Protection API (DPAPI) instead of Base64 encoding
  • DPAPI provides real cryptographic protection tied to the Windows user account
  • Encrypted data can only be decrypted by the same user on the same machine
  • Prevents casual viewing of API keys in the settings.ini file

• Automatic migration from legacy Base64 format (v1.6 and earlier)
  • When retrieving a credential, the system first tries DPAPI decryption
  • If DPAPI fails, it attempts Base64 decode (legacy format)
  • Successfully decoded legacy credentials are automatically re-encrypted with DPAPI
  • Seamless upgrade experience - no user action required

• New DPAPIEncrypt() function using CryptProtectData Windows API
• New DPAPIDecrypt() function using CryptUnprotectData Windows API
• Legacy Base64Decode() retained for migration support only
• Base64Encode() removed (no longer needed for new credentials)

• Menu hover alignment issue - Fixed parallax-like effect where menu highlight would drift from cursor position as you moved down the menu
  • Root cause: Coordinate conversion using simple subtraction didn't account for window borders and DPI scaling
  • Solution: Now uses Windows API ScreenToClient for accurate screen-to-client coordinate conversion
  • Fix applied to both main prompt menu and provider submenu

• CheckHover() now uses DllCall("ScreenToClient") instead of manual mouseX - winX subtraction
• CheckSubmenuClick() updated with same fix
• GetClientRect API used to get accurate client area dimensions

• Beta Tester program - free AI access for beta testers
  • Beta Tester button in About tab for registration
  • Secure server-side API key storage (key never touches client)
  • 80,000 uses limit per beta tester
  • Automatic fallback to own API key when added
• Updated analytics to use new server endpoint format

• New BetaManager class for beta tester registration and proxying
• New BetaProvider for routing requests through server
• ProviderFactory auto-detects beta mode when no own API key exists

• Recommended models auto-selected when API key is provided:
  • OpenAI: gpt-4.1-nano
  • Anthropic: claude-haiku-4-5-20251001
  • Google Gemini: gemini-2.5-flash
• Recommended models marked with ★ symbol in dropdown for easy identification
• Models sorted with recommended model at top of list

• README updated with actual model IDs and auto-selection documentation

• Animated wait cursor during AI processing for visual feedback
• Wait cursor animation files (wait-1.ani, wait-2.ani)

• Menu now appears on the correct monitor in multi-monitor setups
• README updated to reflect actual prompts (removed non-existent "Translate to English")
• README installation instructions clarified - users need to download full ZIP, not just setup file

• Added credits for cursor script contributors (Serenity, iseahound)

• Installer no longer incorrectly sets FirstLaunch=0 (was preventing "install" analytics event)
• Updater now uses WinHTTP with redirect handling for reliable GitHub downloads

• Download mechanism in updater.ahk now matches installer's robust implementation
• Uses ADODB.Stream for binary file handling (fixes COM SafeArray issues)

• Fixed binary download using ADODB.Stream instead of RawWrite (ComObjArray has no "Size" property error)

• Update success confirmation message shows version number after successful upgrade
• Version info included in error messages for easier troubleshooting
• Fixed registry version to use variable instead of hardcoded value

• Installer now uses WinHTTP with proper redirect handling for GitHub downloads
• Tries both tag formats (v1.4.8 and 1.4.8) when downloading updates
• Better error messages showing which URL was attempted

• License agreement in installer - must accept terms before proceeding
• Comprehensive Terms of Use and Disclaimer in application Settings
• Covers liability limitations, AI content warnings, data privacy, acceptable use

• Installer now forcefully closes running instances before upgrading (using taskkill)
• Added validation for downloaded files - checks size > 500KB to catch HTML error pages
• Better error messages when download fails or file is invalid

• Critical: Installer now downloads application files from GitHub when upgrading
• Previously, upgrades failed silently because Setup.exe was alone in temp folder
• Installer detects if running from download location and fetches AITextTools.exe and Uninstall.exe from the matching GitHub release

• Updater now constructs direct download URL as fallback when release assets aren't listed
• Download failures now offer to open releases page for manual download
• Validates downloaded file size to catch failed/empty downloads

• Download update now offers to open releases page if no installer is attached to the release

• "All Models" button in Models tab to refresh available models from provider
• Allows easy switching to newer models when older ones become outdated

• First launch "install" event sent to analytics (distinct from "startup" events)
• Windows security warning instructions in README for non-technical users
• "Get API Key" links for each provider in Models tab
• Step-by-step API key setup instructions in README for non-technical users

• Analytics payload now includes app_name field for multi-app tracking support
• Installer wizard now shows "Bykov-Brett Enterprises" branding
• About tab tagline changed to "./run the revolution."

• Auto-update system with GitHub release checking
• Automatic monthly update check (1st of month at 2pm, or next online day)
• One-click download and install updates from Settings
• Installer upgrade mode (/upgrade flag) for seamless updates
• Anonymous usage analytics for active user tracking (opt-out available)
• UK/US English spelling toggle in General settings
• "Run on Windows startup" option in installer and settings
• "Start minimized to system tray" option
• Custom tray icon when running from source

• Fully standalone distribution - no AutoHotkey installation required
• All executables are self-contained compiled binaries
• Simplified installer that only uses compiled executables
• App shows settings when launched manually, runs silently when auto-started
• Updated Anthropic default model to Claude Sonnet 4
• Gemini API now uses v1beta endpoint for latest models

• Installer "Finish" button showing cancel confirmation
• Uninstaller errors when closing running instances
• Application not launching after installation
• OpenAI API key not connecting (credential storage path issue)
• Anthropic "model not found" error on first connection
• Google Gemini not showing latest models

• Professional installer wizard for easy installation
• Uninstaller wizard accessible from About tab
• Settings now stored in AppData (works with Program Files installation)
• Add/Remove Programs integration

• Improved installation experience for non-technical users
• Settings migration from old location to AppData

• "Access denied" error when installing to Program Files
• Settings persistence issues with restricted folders

• "Ask AI" prompt with custom question dialog
• Dynamic model lists fetched from providers
• Default provider checkbox on Models and Local tabs
• Anthropic Claude provider support
• Google Gemini provider support
• Theme support (follows system dark/light mode)

• Improved provider connection status indicators
• Better error handling for API connections
• Reorganized settings tabs (Models, Local, Prompts)

• "Control is destroyed" error when closing settings during connection
• Model dropdown not updating after provider connection

• Initial release
• OpenAI GPT support
• Ollama local model support
• System tray integration
• Customizable hotkeys (Ctrl+Shift+J, Ctrl+Shift+K)
• Built-in prompts (Fix Grammar, Improve Writing, etc.)
• Custom prompt management
• Secure API key storage using Windows Credential Manager