Changelog

Introduced template path validation for Atoms Cloud: warnings are now logged when a template path is missing, enabling faster diagnosis of configuration issues.

Added a clear command to the template Redis management script for easier cleanup of template-related cache data.

Introduced the sentry_filtered_engineio_invalid_session_total metric to track and monitor filtered invalid session errors.

Fixed root path handling in the MGX thinking template, improving template loading stability.

Addressed “Too many requests” errors under high concurrency, enhancing request handling stability.

Fixed an issue in the frontend notification flow after the version release.

Launched Ads Agent：

Expanded AI capabilities with support for video, audio, text-to-speech, and speech-to-text workflows.

Added PDF analysis tools, plus new Markdown-to-PDF conversion and chart rendering capabilities.

Improved chat theme switching by notifying the frontend of the current theme in real time via WebSocket.

Added locale-aware support for video documents and tags, enabling language-specific filtering and localized tag names.

Fixed an issue where certain configuration/version lookup failures could leave chat tasks stuck and block future messages in the same chat.

Fixed GitHub remote-only branch switching issues to improve branch handling reliability.

Improved image validation for local SVG and raster images, with better error handling and logging.

Theme: Customizable themes and templates for personalized interface styling.

Video i18n Support: Video content now supports multiple languages.

Cloud & AI Pricing Plan - Phase 2: Enhanced pricing plans with more flexible options.

Global Tooltips Style Update: Unified tooltip component styles for consistent interactions.

Affiliate Program Update: Affiliate commission rate adjusted to 10%. All affiliate-related copy updated across the platform.

Added voice input support, including speech-to-text in message inputs on the home and chat pages.

Expanded language support with Traditional Chinese (zh-TW) and Swedish (sv) across frontend and backend.

Added homepage CTA experiments with multilingual template suggestions that can be inserted directly into the prompt box.

Added an Audience & Analytics section entry with navigation into the Growth experience.

Added an expandable/collapsible Atoms Cloud sidebar for backend projects.

Added route navigation dropdown support in App Viewer so users can quickly switch between available app routes.

Added backend support for Gemini text-to-image, and set gemini-2.5-flash-image as the default image generation model.

Added GPT-5.4 model support and related localized labels.

Improved multilingual CTA content with updated translations for English, Spanish, Turkish, and additional supported languages.

Improved Growth setup UX for GA4 and GSC with clearer guidance, privacy messaging, publish hints, blocked-popup fallback flows, and better no-data messaging.

Improved domain cards to show dynamic status labels such as Live and Paused.

Added a new Magic Link sign-up flow with email verification, including email check, send, verify, and resend capabilities for smoother registration.

Launched the new Overview panel after publish, with improved post-publish navigation and state handling.

Added version restore and preview controls directly in the message/version list for faster iteration workflows.

Added image attachment support in chat, allowing uploaded images to be passed through the chat flow more reliably.

Introduced Cloud & AI Wallet / Balance pages and top-up flows, including transaction history, usage breakdown, and balance display across desktop and mobile.

Added Growth block capabilities with deeper GA4 and Google Search Console (GSC) integration, including property/site connection, verification, reporting, and onboarding flows.

Added Claude 4.6 models and updated the default recommended/default model configuration.

Improved chat input responsiveness and first-screen performance with lazy loading, deferred heavy modules, and optimizations around hydration and editor initialization.

Improved wallet UI layout across desktop and mobile, including clearer balances, upgraded toollip content, better spacing, and more polished tables/cards.

Improved Growth block user feedback, prompts, and authorization states to make GA4/GSC setup easier to understand.

Improved support for custom domains by prioritizing preferred/custom domains in more scenarios.

Added version restore for projects, including restore actions, restore progress states, completion messaging, and rollback protection for failed restores.

Added preview version switching so you can switch preview environments between versions more directly.

Introduced wallet payment support, including balance, usage, recharge order access, and instant payment creation.

Added custom domain management enhancements, including domain purchase/connect flows, batch unbind, improved primary domain handling, and better custom-domain support across publish flows.

Added support for 10 new languages across backend and frontend: Arabic, Dutch, French, German, Indonesian, Italian, Brazilian Portuguese, Russian, Turkish, and Vietnamese.

Added a new user profile page in the frontend.

Added mobile file preview improvements for chat blocks, DR content, and action editor workflows.

Added homepage, workspace, and landing page upgrades, including new homepage modules, responsive layouts, and improved content presentation.

Added analytics events for key user actions, including version creation/failure, app viewer readiness, agent mentions, secret flows, and more.

Added version rating feedback in chat, including thumbs up/down and structured feedback options.

Added SEO Specialist Sarah and expanded SEO-related project support.

Improved version restore reliability with locking, stronger validation, better error messages, safer rollback behavior, and long-running operation handling.

Improved chat performance by reducing redundant backend refreshes, caching tab content, and optimizing message scroll/positioning behavior.

Improved first-screen and bundle performance with more lazy loading, better chunking, deferred initialization, and LCP-focused optimizations.

Improved i18n refresh strategy on both frontend and backend to reduce backend pressure and avoid synchronized refresh spikes.

Improved workspace and editor experience by simplifying component structure and making file/task interactions more consistent.

Introduced custom domain support for apps, including domain binding, DNS/SSL checks, primary domain management, subdomain support, and related user tooling.

Added backend integrations for app configuration and secrets management, including improved environment variable syncing for deployed apps.

Added Stripe connection guidance and improved Stripe/Supabase integration flows for app setup.

Added AI capability features in the frontend, including model library, usage overview, model tags in plans/tasks, and richer model metadata.

Introduced auto model-selection experiments and expanded model options, including new Gemini, DeepSeek, GLM, and image-capable model entries.

Added a MetaGPT landing page and a new Atoms landing/branding experience.

Added support for larger Max plan storage, with Max plan capacity increased from 40GB to 100GB.

Added support for bucket limits in App Storage, including a maximum of 10 buckets per user.

Improved security around uploads, route handling, and rendered content, including stronger filename sanitization and safer HTML handling.

Improved storage and quota handling across plans, including clearer quota enforcement and larger Max-tier allocations.

Improved AI model presentation and organization with richer labels, descriptions, icons, and categorization across languages.

Improved landing page, pricing, and header/footer experiences.

Human-in-the-Loop

Human-in-the-Loop is a new review step that ensures MGX Agents fully understand your intent. It presents a clear to-do list you can edit by typing #. After you approve it, the agents execute the plan with greater accuracy.

Power Up with Deepseek 3.2 Exp

The DeepSeek 3.2 Exp model is now available. Enjoy faster reasoning, stronger execution, and improved reliability across your projects.

New Edit Mode in the Editor

A dedicated Edit Mode is now available in the Editor, giving you finer control and a smoother experience when refining content.

Prototype Shortcut

Prototype is now accessible directly from the main page for faster iteration.

AI Feedback Assistant & Help Center Link

Using the Feedback button (as shown in the picture below) on any message now opens a guided assistant to help you resolve issues. Additionally, a direct link to the Help Center is now available in your profile menu and Settings.

Enhanced Slides: Improved visual quality and layout for all generated slides.

Faster Page Rendering: Wiki and Video pages now load noticeably faster.

Quick Sharing: Deep Research articles now auto-generate clearer, more descriptive share titles.

Platform Stability: Resolved multiple known issues to deliver a smoother, more reliable experience.

AI Image Generation (Nano Banana Model)

Added support for automatic, context-aware image generation. The Nano Banana model enables Alex (The Engineer Agent) to produce consistent, high-quality visuals that match website content and design style.

Introducing How-to Guides (Creator Playbook)

The new How-to Guides page provides step-by-step tutorials on building various projects with MGX, such as high-converting SaaS landing pages, calendar & scheduling projects, and an English learning platform.

Explore more possibilities on How-to Guides: https://mgx.dev/how-to-guides

Better Downloads: Resolved garbled text in Deep Report downloads and enhanced overall download reliability.

Clearer Credit Usage: Redesigned the free credit usage bar for better clarity, so you can easily track your monthly credits.

Smoother UI/UX: Improved key interactions, including Deep Research report downloads, in-chat message sending, and block display logic.

Platform Stability: Fixed multiple known bugs to enhance overall performance and reliability.

Introducing the Video Center

In Video Center, users can find official tutorials, in-depth guides, and practical case studies. It helps you learn more about MGX's powerful features and quickly start building.

Discover the App World Showcase

We have added a Showcase tab in App World. It helps users to explore the potential of MGX with a selected group of high-quality projects.

Enhanced Account Security

We've strengthened our password requirements to include special characters.

Instant Connection Alerts

If the Supabase connection is disabled, you'll now receive a clear, instant notification during a chat.

Smoother UI/UX: Enhanced expand and collapse behavior for the “Task for human” input box. Updated visual indicators for files that can’t be previewed.

Improved Security Logic: Added clearer expiration details and retry limits to email verification and password reset links to prevent misuse.

Clearer Billing: Clarified billing and deduction logic for users upgrading from an annual plan to a higher-tier monthly plan.

Platform Stability: Updated backend logic for the Affiliate Program and fixed multiple bugs to improve overall platform performance.

Annual Subscription Plan

We’ve introduced an annual subscription option to provide users with greater cost efficiency for long-term projects. When purchasing the first-tier credits in the Pro or Max plan, you’ll receive a 21% discount. For all other tiers, the discount rate is 18%.

For details about pricing, billing, and credit allocation, refer to the MGX Annual Payment Plan.

Brand New Use Cases Page

We’ve redesigned our use cases page to better communicate our mission of bringing ideas to life with AI Agents. The new layout delivers a cleaner look, clearer messaging, and a more intuitive start for new users.

Language Switching

We've added four official languages, including:

Users can switch languages from the top-right menu.

Replay

Replay in App World — You can now watch the full step-by-step creation process of any app directly in App World. The new Replay feature reveals how complex projects are built from a single prompt.

Explorer Program Launched

Join the MGX Explorer Program to gain early access to major beta versions, verified identity status, monthly rewards, and official exposure opportunities. Explorers can contribute by engaging with the community, sharing feedback, and promoting MGX across platforms. Apply now to shape the future of AI collaboration.

Affiliate Program Launched

You can now earn commissions by referring new users to MGX: 30% commission on each referred user’s first six payments + 10 credits for your friend.

For more details, see Affiliate Terms of Service

Default Engineer Mode

Workspaces now default to Engineer Mode for a more powerful initial experience.

UI/UX Improvements

Enhanced wording for Supabase connection prompts and improved visual effects for slides.

Bug Fixes

Resolved multiple known issues, significantly improving platform stability and performance.

Fully refreshed product design.

Simplified pricing plans: Free, Pro, and Max. Pro and Max can be paired with different credit tiers.

Launched a new AI agent, Iris, the Deep Researcher, supporting in-depth research scenarios.

Introduced BON (Race Mode).

Added semantic Actions.

Rebuilt the PPT creation flow with improved prompts and outputs.

Released new models: GPT-5, Gemini 2.5 Pro, Qwen3, and DeepSeek 3.1.

Updated sidebar layout and added “star” for chat lists.

Publish settings are now integrated with Domain in the Settings panel.

Added support for connecting Supabase projects directly from the homepage.

Paid users can now set default chat permissions (e.g., default private) in Settings.

Improved Remix display format.

Optimized model list presentation.

Enhanced quick-access interactions.

Fixed several known bugs.