* Add: Cross Site now support shorcuts (Quick Actions), blocks (GDPR-friendly), etc.
* Update: Modernized all Gutenberg Blocks.
* Fix: Avoid crashes caused by Claude when using MCP tools together with function calling.
* Fix: Prevent React-related crashes during streaming.
* Add: Support for the Claude Sonnet 4.6 model.
* Add: Forward Anthropic keepalive events to the browser.
* Add: Make Realtime transcript captions clickable links.
* Add: Introduced role-based access control for MCP with three levels (admin, read-write, read-only).
* Add: Improved Content Aware features (better placeholder replacement, more accurate context detection, etc).
* Fix: Automatic bearer token generation when MCP is first enabled.
* Update: Improvements in the Realtime module, more to come soon.
* Update: OpenAI now exclusively uses the Responses API, simplified and optimized the code.
* Add: Support for the new Claude Opus 4.6 and Opus 4.5 models for improved AI responses.
* Update: Deprecated GPT-4o, GPT-4o mini, GPT-4.1, GPT-4.1 mini, and their related realtime/transcribe models to guide users toward newer, supported options.
* Update: Replaced shortcut-based data storage with AES-256 encryption to avoid cache corruption issues.
* Add: Enhanced the wp_update_post MCP tool with better truncated JSON detection, content verification, and richer response data.
* Update: Linked PDF chunks via unique refChecksum and added ref filter to embeddings table.
* Add: Added WooCommerce MCP module with 25 tools for managing products, orders, inventory, customers, reviews, and analytics.
* Fix: Prevent duplicate shortcut transients from being created on every page.
* Fix: Hotfix TypeError crashes in MCP tool calls when JSON-RPC requests are missing an id field.
* Add: Introduce a new wp_upload_request MCP tool that generates one-time upload URLs so agents can upload local files via curl instead of base64.
* Add: Add a new MCP "Database" module that lets you execute SQL queries.
* Update: Remove the default popup title and AI name fallback so leaving the title field empty now shows no title at all.
* Fix: Resolve a PHP 8.2 deprecation warning related to a dynamic property.
* Add: Introduce a Streamable HTTP transport for MCP so Claude Code can connect directly to WordPress.
* Fix: Prevent unsafe file renaming during metadata updates.
* Fix: Hotfix SSRF vulnerabilities by using wp_safe_remote_get for external URLs and secure URL fetching in audio transcription and file handling.
* Add: Improved embedding dimension handling with matryoshka support, showing fixed vs configurable dimensions in the UI with clearer tooltips.
* Update: Redesign and refine the ChatGPT Theme and Foundation Theme with improved layouts, bubble styling, and more flexible avatar/name display and header titles.
* Add: Introduce 'context framing' for embeddings with a new filter.
* Add: Moved Embeddings "Sync Settings" into Settings > Knowledge with separate Local Content and Remote URLs sections and a new Sync Interval option.
* Add: Store Shortcuts prompts securely on the server and only send a shortcutId to the client so prompt text is no longer exposed.
* Add: Load plugin translations properly with load_plugin_textdomain to ensure all text is translatable.
* Fix: Via MCP, ensure scheduled posts created via wp_update_post respect the schedule_for parameter and edit_date flag so they publish at the correct time.
* Add: New MCP integration for Polylang.
* Fix: Improve Gemini 3 support, restoring vision capabilities, function calling, and enforcing a single candidate response, and fixed image generation pricing.
* Add: Support image generation via OpenRouter.
* Fix: Resolve undefined array key warnings in Remote URL sync and only update timestamps when content actually changes.
* 🎵 Discuss with others about Ai Engine on [the Discord](https://discord.gg/bHDGh38).
* 🌴 Keep us motivated with [a little review here](https://wordpress.org/support/plugin/ai-engine/reviews/). Thank you!
* 🥰 If you want to help us, check our [Patreon](https://www.patreon.com/meowapps). Thank you!
* 🚀 [Click here](https://trello.com/b/8U9SdiMy/ai-engine-feature-requests) to vote for the features you want the most.
* Add: New "Remote URL" feature for Embeddings to create and auto-sync embeddings from external URLs with change detection.
* Fix: Corrected auto-scroll behavior for non-streaming chatbots so conversations stay in view.
* Add: Enabled multi-block support for Magic Wand tools (Correct Text, Enhance Text, and Translate Text).
* Add: Introduce a user.query filter in the JS API so developers can intercept, modify, or cancel user messages before sending.
* Add: Support next_run_delay for multi-step tasks to better control task scheduling.
* Fix: Resolved an issue where multi-file uploads were empty.
* Fix: Ensured the Standard Header correctly displays the Popup Title.
* Fix: Adjusted link colors in the Timeless theme.
* Fix: Prevent shortcuts from bypassing locked chatbots (GDPR consent, for example).
* ⚠️ Important: Changed the mwai_chatbot_reply filter to pass a reply object instead of a query. Developer! Use reply->query where query was previously used.
* Update: Removed unused files and code to reduce clutter and improve performance.
* Info: Wishing you a Happy New Year 2026 ☀️🎉 No doubt this year will be full of surprises again, let’s continue riding the AI wave together! 😊
* Add: Added title filter to the Embeddings table.
* Add: Added filter to all model dropdowns.
* Add: Enabled custom placeholder support and model-aware credit customization via new and updated filters for developers.
* Add: Introduced prompt caching for Anthropic and a free-text dimension input for non-OpenAI embeddings.
* Add: Added dimension mismatch warnings for Pinecone and Qdrant to help diagnose configuration issues.
* Add: Included "speech_recognition" support in the embed.js system.
* Fix: Resolved multiple embedding issues, including OpenRouter parameter compatibility, Pinecone Serverless URLs, and vector store file uploads.
* Fix: Improved compatibility with multilingual plugins, placeholder processing, and post content handling through dedicated filters.
* Update: Performed an end-of-year cleanup by removing expired database migrations, deprecating outdated Claude 3 Opus models, and tidying up legacy debug and commented code.
* Add: Added support for the new gpt-image-1.5 model, works quite well! 😊
* Fix: Restored reliable image editing (added missing file attachment handling, separating the edit mode prompt from templates, and preventing double dots in generated filenames).
* Fix: Corrected Mistral model filtering so available models now appear and work as expected.
* Fix: Resolved a UI crash when selecting OpenRouter providers that use non-standard model data.