- New JS API features: - Support creating/editing/deleting character cards - Support creating/editing/deleting user personas - Support creating/editing/deleting world books - Support setting the current chat name, character, user persona, and world book - Support reading and writing long-term memory - Support initiating AI generation requests - Support reading/writing/sending input box content - Support exporting/downloading files - API parameter settings moved to the top-right corner with added visual configuration - Added Frequency Penalty and Presence Penalty to API parameter settings - Optimized JanitorAI URL import process - Presets now support item duplication - Added top navigation for model selection in OpenRouter & Volink - Fixed output disorder during streaming caused by regex - Fixed global variable saving issue - Fixed request failures caused by OpenAI max_tokens deprecation - Removed some deprecated models - UI detail improvements - Known bug fixes

- TTS API Optimizations: - Added OpenAI support - Added support for custom OpenAI-compatible protocols - Supports searching or directly entering Model IDs - New Creator Tools (Enable in Chat Settings -> Creator): - Variable Manager: View or modify variables - JS Console: View JavaScript logs for debugging - Context Log: Now relocated to the "Creator" section - OpenRouter & Volink model selection now displays context window and pricing - Added support for importing chat history from Saucepan - Added support for importing presets from Chub - Macro Optimizations: - Fixed issue where random number macros generated identical values - Fixed timezone issue with {{time}} - Fixed issue where setting JSON within `setvar` macro parameters was ineffective - Added macro escaping support, e.g., `\{\{char\}\}` will render as `{{char}}` - Optimized the style of shortcuts buttons in Group Chats - Optimized image caching - Fixed issue where JavaScript could not write to the input field - Minor UI refinements - General bug fixes

- Added support for Claude Opus 4.6 with 1M context - Added fullscreen & macro assistant toggle to chat shortcuts - Added auto-capitalization and auto-correction for Latin-based input methods - Unified macro naming, keep old version compatibility - Optimized chat list loading performance under extreme conditions - Fixed an issue where variables were not cleared during chat restart - Fixed an issue where long-term memory was not cleared during group chat restart - Fixed a Markdown bold text display issue - Fixed an issue where early macro tags could fail in certain cases - Fixed a JavaScript rendering issue when rewriting messages - Fixed a line spacing issue after toggling JavaScript on some devices - Fixed an issue where JavaScript could render the variable table in certain cases - Fixed an occasional issue where `tavo.set` in JavaScript would fail - Minor UI optimizations - Fixed known bugs

- Macro:** Added support for global variables - JS Support: Added variable storage API (refer to the official documentation for usage instructions) - JS Support: Improved visual effects for pop-up dialogs - JS Support: Fixed issue where copying to clipboard could not be triggered - Optimized regex rendering performance for stream output - Optimized exported file name suffix format (now using a shorter format) - Fixed import issue with new version group chat JSONL files - Added Spanish language support! - Private chats and group chats can now be switched seamlessly - In full-screen editing, tapping back now saves automatically - Regex entries now support copying - User identities now support search - Chat import now supports the Agnai platform - Chat variable data is now exported together with chats - The long-deprecated “Non-standard HTML tag display” option has been hidden and will be removed soon - Fixed an issue where message content on the right side was left-aligned in tablet mode - Fixed bubble height and width calculation issues when JS is enabled - Extensive popup UI optimizations - UI detail improvements - Various known bug fixes

- **JavaScript support!** Two modes: script mode and codeblock mode - Enhanced security in advanced rendering - Adjusted settings menu structure: - Display and Other Settings merged into Chat Settings - Image Sending moved to Chat Settings - Some items moved to "About Tavo" - Added a standalone settings page for **Advanced Rendering** - Smart recognition of imported file types with correct import handling - Fixed occasional character and chat history disorder issues during chats - Deleting a character now also removes associated lorebooks and regex - Optimized avatar loading performance, resolving lag caused by oversized avatars - Optimized database and increased the maximum limit to 3GB - When importing files exported by Tavo, prefixes and suffixes are now automatically removed from names - Fixed CSS issues that caused rendering to stop in non-advanced rendering modes - Resolved message loading timeout issues under complex regex conditions - Fixed ordering issues with presets and worldbooks at the same injection depth - Fixed known bugs

- Optimized the character card preview and chats page - Added preset search functionality - Added regular backup reminder (default: 14 days, customizable) - Added macro assistant for quick access during input - Supported importing wyvern chat logs - The left sidebar now displays input field drafts - Optimized the execution frequency of RegEx during streaming to improve performance - Editing characters, lorebooks, etc., no longer returns to the top of the list - Update the default English font in Advanced Rendering - Fixed the issue with {{char}} references in group chat opening lines - Fixed the issue where some devices were unable to select images. - Fixed crashes caused by avatars on certain devices - Fixed English font issues in abnormal speech bubbles - Minor UI optimizations

- Support sending images to AI! (Need to enable in settings) - Added shortcut speaking in group chats - Clicking on character avatars now previews the full image - Claude and Grok now use new API, supporting more custom parameters - Fixed issue of persistent loading (infinite loading) - Fixed memory overload when switching chats caused by overly large character avatars - Hidden bubbles are now excluded from depth calculation in regex - Slightly optimized the color algorithm for input box - Fixed issue where dividers appeared only at the bottom in flat style - Fixed issue where top buttons were blocked in immersive mode - UI detail optimizations - Known bug fixes

- Character, preset, and other lists now support drag-and-drop reordering (pin feature removed) - World Info now supports entry search - Regex replacement text now supports macro injection - Advanced Rendering: slight upward swipe during streaming prevents auto-scroll - Stop button now correctly halts group chats - Copying chats now includes variables, regex, Lorebook state, and Quick Switch - Optimized official "Viridian" theme (please copy before updating if you wish to keep the old version) - Optimized Chat History construction - Fixed lorebook depth calculation errors causing injection failures - Fixed occasional preset save failures - Fixed issue where chats occasionally became empty under advanced rendering - Fixed issue with duplicate selection boxes appearing on some bubbles during multi-select - Fixed issue with deleting second greeting message - Minor UI adjustments - Bug fixes

- API now allows custom parameters (request body) - Added character filtering to the chat list - Added font size adjustment in quick-switch - Added "Select to here" in advanced rendering - Fixed issues with dragging presets, lorebooks, and regex across screens - Fixed the issue where only 20 messages could be selected in advanced rendering - Fixed the issue with selecting APNG as avatars - Fixed the issue with editing default persona - Fixed compatibility issues with lorebook imports

- Added new model: Gemini 3.0 Flash Preview - Refactored advanced frontend rendering to improve performance and compatibility, resolving slow loading and blank screen issues - Group chats now support specifying a character opening message - Fixed an issue where some models were not displayed in OpenRouter - Fixed an issue where hiding the model’s chain-of-thought did not work under advanced frontend rendering - Fixed an issue where copied messages were not the original content under advanced frontend rendering - UI detail optimizations - Known bug fixes

- Support Character Card Spec V3: - Extra fields nickname and group only greetings - Added "source" field to enable lightweight copyright protection (prevents removing other authors’ sources) - Support importing CHARX and APNG (animated avatar & background) - Card creation and modification timestamps - Support theme import/export - Rename button in the left sidebar - Output interruption - Chat input box now supports full-screen mode - New models: - DeepSeek 3.2 Stable - Claude Opus 4.5 (200K & 1M) - Grok 4.1 - GPT-5.2 series - Model settings now allow temperature, top-p, and top-k input - Fixed an issue where advanced rendering caused Gaussian blur - Fixed conflict between temperature and top-p in Claude - Fixed editing issue after regeneration - Fixed lorebook cooldown issue - UI detail optimizations - Bug fixes

- Added support for the Gemini 3.0 Pro Preview model - World Book and Regex now support pinning to top - Supports hiding the model’s reasoning process, toggleable in Settings > Display - Character list now supports grid view - Storage now allows cleaning invalid character cards and related data - TTS voice now supports background playback, toggleable in Voice > Voice Settings - Visual Novel mode has been moved to Theme > Bubble > List Style - Improved UI interactions in model settings - Fixed lag issues when receiving messages - Advanced Rendering now supports one-tap scroll to top/bottom - Advanced Rendering performance and animation optimizations - Advanced Rendering regex-related fixes - UI interaction optimizations - Bug fixes

- Experimental feature: Advanced frontend rendering! Greatly enhanced HTML support in chat bubbles! (Enable it in Settings - Display) - Support importing chat history in chub format - Background character image switching in group chats now syncs with voice playback - Optimized text rules for TTS playback - Improved storage space management - Fixed issues with long memories when upgrading data manually - Interaction details optimized - Bug fixes

- Added storage management, allowing users to clear cache, logs, and optimize the data - Restored the explicit 'Scroll to Top/Bottom' buttons (shown when scrolling) - Optimized the performance of the 'Scroll to Top/Bottom' feature - Improved World Book injection performance to reduce lag when sending messages - Fixed various display issues after regenerating - Fixed the TTS 'Play Quoted Only' feature - Fixed the GLM returning 400 issue - Fixed bubble display issues caused by regex in 'on receive' - Minor UI improvements

- Added Grok support - When importing a character card with the same name, you can now choose to update or save as new - Added fold feature by long-pressing chat bubbles - Added quick scroll-to-top/bottom buttons while chatting - Added protection against accidental multiple message sends - Refactored chat code for improved rendering performance - Disabled Top K by default - Optimized scroll positioning speed from search results to bubbles - Fixed issue where some HTML tag styles were not applied - Fixed `{{char}}` not working when rewriting user messages - Fixed TTS playback issue with italic text - Fixed companion days display issue after importing chats - Fixed character avatar issue when importing from Chub - Fixed possible white screen when deleting messages - Fixed occasional white screen issue during search - Fixed data cleanup issue - Minor UI improvements

- Bug fixes

- Added chat history search! (Search icon on the right side of the sidebar) - Support for Claude Sonnet 4.5 - Fixed an issue where duplicate APIs appeared in load balancing - Fixed an issue where calling Grok through third-party services failed - Fixed an issue where excessive data caused backup failures - Fixed a backup failure caused by empty character card images - Fixed export failures under certain filenames - Fixed avatar misalignment issue in tablet bubble style - Optimized UI details - Bug fixes

- Long-term memory can now be configured with a separate API - Added configuration for memory injection positions to enhance memory effects across different presets - Optimized default prompts for long-term memory to reduce chain-of-thought and format extraction - Added long-term memory macro {{summary}} compatible with Tavern - Fixed backup failure issue - Error popups now allow copying full error details - UI detail optimizations - Bug fixes

- Regenerate now allows keeping one selected message from multiple - Group chats switch background based on speaker role - Chat optimizations - Switching chats while replies are loading no longer loses messages - Original message is no longer lost when regeneration fails - Buttons at the bottom of bubbles will still show when errors occur - Red error bar will appear if AI Assist (Lingxi) fails - Multi-select bubbles now support "Select All" - Quick view mode now supports one-click scroll to top/bottom - Rewriting now uses custom shortcuts - Load balancing now only keeps the latest context log - allows setting Enter to send (enable in other settings) - Fixed some markdown rendering issues - Optimized UI details - Bug fixes

- Now you can create Load Balancing on the API page to connect multiple APIs: - Provides multiple strategies such as round-robin, random, and weighted - When an API fails, it will automatically switch to the next retry; after reaching the retry limit, the last error will be returned - If authentication fails, the API will be disabled and can be manually restored - Provides real-time logs and statistics - Allows setting retry count and log size - TTS API can now be copied - UI interaction optimization - Bug fixes

- Added customizable chat shortcuts - New “Wisteria” and “Theatrical” themes - Visual Novel mode now available with themes & sidebar quick switch - Optimized bubble rendering (Markdown & HTML) and performance with large data - Chat history export to TXT supported - Character cards can be imported from gallery - Added system TTS using your device’s voice engine - Light mode is here! Switch in Settings → Appearance - Improved backup & restore reliability - Enhanced UI details & interactions - Bug fixes and performance improvements

· Added support for tablet-style message layout, ideal for reading novel-like content! You can enable it via Theme > Bubble > Message Style · Improved the model settings page interaction — now temperature, top-p, and top-k can be disabled (they won't be sent to the model and will use the model's default values) · To prevent response truncation, the reply token limit is now unlimited by default · Fixed an issue when matching a large number of groups in regex · Minor UI improvements · Bug fixes

· Added TTS voice support · The option “Show non-standard HTML tags” is now disabled by default. You can enable it in “More” > “Display” if needed · Fixed import failure issues with non-standard character cards · Fixed compatibility issues with importing certain non-standard character cards, presets, world books, and regular expressions · UI interaction optimized

· Fixed the issue where hidden messages were not working · Added hidden message parsing when importing/exporting chat logs · Fixed occasional segment loss during streaming output · Fixed an issue where messages were not replaced by regex in certain group chat scenarios · Minor UI improvements

· Added backup feature! Can be used for device migration and full data backup! · Character cards now handle associated world books and rules when importing and exporting! · World books now support multi-instance. · Added hide floor function, which can erase messages from AI memory without deletion. · AI now supports real-world time awareness, referencing the official preset in "Light Strings" "⏰ Real-world time". · Added "Non-standard HTML tag" display switch, for a closer tavern experience when turned off. · Model settings page added top k switch, forcing no top k parameters sent to model when turned off, addressing issues with intermediary service providers not handling parameters for different models. · iOS official themes now default to no Gaussian blur, customizable by preference. · Fixed issue where long-term memory was retained after session restart. · UI optimization.