What's New

• Subscriptions through Paddle — Basic is $5/month or $50/year. Pro is $15/month or $150/year. Manage your subscription and receipts from billing settings.
• Monthly AI credits — Free includes 50 credits, Basic includes 300, and Pro includes 1,000 each month. See an estimate and maximum charge before each hosted AI run. Annual plans also receive credits monthly.
• More model choices — Use Gemma 4 on Free, GLM-5.3 Flash on Basic, and GLM-5.3 or DeepSeek V4 Pro on Pro. Pro web research uses the same credit balance.
• Review your plan before you build — Pro can check requirements and technical plans for gaps and conflicts. Read each suggested correction and choose which ones to apply.
• Cloud tools for longer projects — Basic supports 15 cloud projects and Pro supports 50. Pro also adds named cloud versions, comparison and restore, and reusable private templates.
• Keep the full workflow free — All five planning stages, exports, unlimited local projects, and local version history remain available on Free. You can also use your own AI provider key.
• Better buttons and page spacing — Pricing controls now have clear selected states, complete card borders, and matching button heights. Public pages have less empty space below the header.
• Smoother navigation — A small loading indicator and a short fade replace the large page-loading spinner. The desktop background animation is lighter, with a static background on mobile and when reduced motion is enabled.
• Small fixes that matter — Refreshed favicon links, improved save-status contrast, clearer screen-reader labels, and corrected dashboard link targets.

• Focused Question-by-Question Flow — Step 1 (Research), Step 2 (PRD), and Step 3 (Tech Design) now guide users through focused question cards with 1-click suggested options, keyboard shortcuts (⌘↵), progress dots, and a View All Fields switch.
• Zero-Account External AI Workflow — Added dedicated Copy Prompt and Paste-Back modes across Steps 1 to 4, enabling users to generate prompts for ChatGPT, Claude, or Perplexity and continue the pipeline with zero logins or API keys.
• Dismissible Login Prompts — Users can now easily dismiss in-app sign-in prompts across all 5 steps to keep the workspace focused and clean.
• Instant Project Kit Export — Added a prominent quick-export banner to the top of Step 5 so you can download your complete 12+ file project kit ZIP or copy all markdown immediately.
• Unblocked App Idea Entry — Removed minimum character restrictions on initial app ideas so you can start planning with ideas of any length.
• Refined Overlay Layouts — Moved cookie consent and toast dialogs away from navigation action areas to prevent button obstructions.

• A calmer workspace from start to finish — The dashboard, settings, all five planning steps, mobile navigation, loading states, dialogs, provider controls, and project forms now use clearer hierarchy, flatter surfaces, larger touch targets, and more consistent actions.
• A clearer planning-to-execution journey — The homepage and supporting public pages now explain what each stage produces, how the artifacts work together, and what a typical end-to-end plan may cost across major AI models.
• Modern provider and model controls — Provider setup, model selection, saved-model migration, pricing context, advanced options, and runtime routing now stay aligned across the workspace.
• Research choices that explain themselves — Standard and Deep Research now make their capabilities, search behavior, expected depth, and usage tradeoffs easier to compare before generation.
• Smarter Step 5 execution — Autonomous, multi-agent, and phase-by-phase workflows now explain when to use each mode, preserve the selected architecture, and require verification plus a useful handoff.
• Dynamic Build Plan phases — Guided execution reads the real phases from each generated Build Plan instead of imposing a generic backend, UI, and API sequence.
• Recoverable agent handoffs — Goal and teamwork prompts include the correct kickoff commands, execution reports can be brought back into the workspace, and partial progress remains understandable after interruptions.
• A complete local Export Kit — ZIP exports include project documents, agent configuration, launch protocols, generated execution prompts, and a clearer extract-open-run guide. ZIP creation remains a local final step.
• GitHub export that starts cleanly — GitHub connection now uses the normal sign-in flow, gives clearer setup feedback, and avoids the previously disabled manual-linking path.
• A dependable demo and cleaner cloud sync — Explore Demo reliably opens a ready-to-use example, while bundled demo projects stay local instead of producing invalid cloud-sync requests.
• Privacy-first sharing — Project preview images exclude descriptions, documents, generated files, and project IDs, with native sharing plus dependable image, download, and caption fallbacks.
• More resilient generation — Typed validation, streaming states, cancellation, retry behavior, incomplete-draft recovery, and provider errors are handled without hiding useful existing content.
• Production-ready foundations — Authentication, project persistence, quotas, analytics, security policies, and production setup are documented and covered by expanded workspace, accessibility, security, and browser tests.

• More room to build - Basic and Pro plans now unlock higher weekly usage, more projects, and more step adjustments.
• Pro web search - Pro users can use a stronger search-enabled model for research-heavy prompts and product planning.
• Clearer quota status - Usage now explains the limits that apply to your plan instead of showing a generic counter.
• More reliable generation - Streaming, model switching, and required setup fields are more stable across the workflow.
• Cleaner workspace pages - The research, PRD, tech design, agent config, and delivery steps now give more room to the current task and keep generated output easier to read.
• Better public pages - The landing, examples, how-it-works, and template pages now use clearer copy and a more polished layout.
• Advanced telemetry on demand - The telemetry console is now tucked into Advanced settings and can be dismissed when you do not need it.

• Faster return to work — Returning users now land directly in their workspace instead of being sent back through the public page.
• Smoother dialogs and drawers — Settings, welcome screens, and overlays now behave more consistently without background scroll issues.
• Guided first run — New users get a clearer welcome flow for the 5-step vibe coding pipeline.
• Safer AI connections — Bring-your-own-key requests for hosted providers now run through a secure backend path, while local providers stay local.
• Privacy-first analytics — Analytics and speed monitoring stay off until you opt in through cookie preferences.
• Faster loading — Heavy services load only when needed, reducing the first page payload.
• Instant settings sync — Preferences, API keys, and model parameters update across the wizard without a refresh.
• Sharper interface polish — Improved contrast, focus states, and visual treatment across the workspace.
• Clearer search availability — Free AI now explains when live search is unavailable and points you toward search-capable providers.

• Unified Step 5 Workspace — Consolidated the build plan, phase prompting, and export kit into a single interactive three-tab workspace, replacing the separate build and export pages.
• Antigravity Suite Upgrade — Renamed Gemini CLI to Antigravity CLI (invoked via agy), and split Google Antigravity into the Antigravity 2.0 desktop app and the editor-integrated Antigravity IDE.
• Corrected Setup Instructions — Documented Antigravity 2.0 as a standalone desktop app (instead of npm) and updated Antigravity CLI to use curl/powershell scripts with browser logins.
• Native Codex Support — Added native support for Codex with TOML-style rule outputs (.codex/rules/default.rules).
• Complete Kickoff Prompts — Configured prompt templates to explicitly instruct agents to read all context documents and execute goals from start to finish rather than just starting Phase 1.
• Performance & Cleanup — Lazy-loaded the syntax highlighter to optimize LCP, excluded macOS metadata (__MACOSX) from ZIP exports, and removed redundant build/export views.

• Need more usage? — A NVIDIA option now appears right below Gemma 4 so users can switch to their own NVIDIA access instead of waiting for the daily reset.
• Free vs BYOK split — Settings and the API key modal now separate Gemma 4 (Free AI) from bring-your-own-key providers.

• Gemma 4 (Free AI) — Added a no-key AI option for one daily project and five adjustments per step.
• Daily safeguards — Usage resets at 00:00 UTC, with export remaining local and free.
• Deep Research model selector — Gemini Deep Research Max and standard Deep Research now show consistently across all personas.

• Gemini 3.1 Pro Agents — Added Deep Research Max for comprehensive analysis and updated the standard Deep Research model.
• Pricing Transparency — Cost estimates based on research depth (~$1.00-$3.00 for standard, ~$3.00-$7.00 for Max) are now shown directly in the UI.
• Extended Limits — Agentic deep research workflows can now run up to 60 minutes for intensive market gap discoveries.

• 6 new providers — xAI (Grok), DeepSeek, Groq, Mistral, and Qwen now available alongside Gemini, OpenAI, Claude, OpenRouter, and Ollama
• Tailored controls for each provider — Grok can now search the web, DeepSeek has a thinking mode, Groq and Grok have reasoning effort sliders, and Mistral has a safety toggle
• Advanced settings adapt automatically — the settings panel only shows options your chosen provider actually supports
• Better generation tuning — new sliders for repetition control and sampling across all compatible providers

• AI responses now stream through Vercel's serverless functions — your API key is sent over HTTPS, used for that single request, and never stored or logged
• Unified streaming engine under the hood — all 10 providers use the same reliable pipeline
• Cleaner, faster experience with less code running in your browser

• Fixed a rare error that could occur after app updates (GitHub issue #10)
• App updates now apply automatically in the background
• Strengthened security headers and removed unnecessary build-time key exposure
• Reduced deployment size for faster loading
• Updated default Gemini models to the latest versions (Flash, Pro, Flash Lite)

• Added OpenAI-Compatible provider: connect to Groq, Ollama, LM Studio, or any OpenAI-compatible endpoint
• Realtime WebSocket streaming support for compatible endpoints
• New post-connection success screen showing provider name, default model, and a shortcut to customize model settings
• Actionable toast notifications with contextual error messages for API failures
• Connection testing with clear error feedback before saving

• Fixed API key modal incorrectly showing for users with non-Gemini provider keys
• Added Vercel Streamdown for improved markdown/code/math rendering
• Added new shadcn UI components (button, button-group, separator, tooltip)
• Updated design tokens and styling

Added a dedicated "What's New" page to display project changelog and recent updates in a beautiful timeline format.

Improved API key validation and input experience in settings. Refined the VibeCoder research flow to pre-fill project details automatically, reducing manual setup time.

• Updated AI model names to their official product names across the app
• Fixed persona routing issues
• Introduced a content-rich landing page with improved app route isolation

• Added model pricing display so users can estimate costs before generating
• Enhanced the AI generation button with better feedback states
• Implemented provider-specific model selection
• Expanded advanced settings including Gemini's thinking budget and web search grounding
• Improved mobile UI/UX across the board

Added npx commands for one-click skill installation. Integrated installation commands into skill data, detail pages, the skills index, and the installation guide.

• Major SEO enhancements with additional programmatic SEO pages
• Created About page
• Added multiple blog posts covering AI coding documentation, Claude skills, coding workflows, technical PRDs, and more

• Introduced new CSS theme with cyberpunk design tokens
• Optimized UI animations, background rendering, and font loading for improved INP, LCP, and battery life

• Introduced a reusable AI loading spinner component with smooth animations
• Added cost warning for Gemini deep research agentic workflow
• Minor formatting adjustments throughout

• Mobile step navigation with swipe gestures and floating action button
• Swipe-to-dismiss for modals on mobile
• Touch-friendly scrollbars and tooltip positioning
• Mobile file viewer for wizard pages
• Transform card for mobile external AI workflow
• Collapsible thinking status card with universal mobile cancel
• Added Vercel serverless proxy for Deep Research to bypass CORS
• Introduced "How It Works" page with enhanced SEO

• Built the complete 5-step wizard UI (Research → PRD → Tech Design → Agent Config → Export)
• Implemented WelcomeModal, Home page, and full Layout component
• Added p5.js flow field animated background with reduced motion support
• Created Terms of Service, Privacy Policy, Contact, and Security pages
• Introduced Vibe Workflow skills system with dedicated pages and data
• Added optimistic updates for project data with new schemas
• Implemented project context extraction from PRD and tech outputs
• SEO meta descriptions, JSON-LD structured data, and persona landing pages
• Claude skills for workflow, agents, and build steps

• Added Security page explaining client-side API key management
• Implemented Content Security Policy headers
• Created SEO components and pages for different user personas (VibeCoder, Developer, Learner)
• Updated copyright year across the app
• Integrated AI agent skills for Vercel React best practices

Added core application constants, model configurations for all supported AI providers, and tool definitions that power the agent configuration step.

• Implemented full PWA support with service worker, manifest, and update prompts
• Added build chunking for vendor libraries and PWA cache size limits
• Added MIT license
• Updated README with multi-provider AI support details, advanced settings, and tech stack
• Added privacy and security documentation for API key handling

• Added all project workflow pages: Research, PRD, Tech, Agent, Build, Export, Projects, and Settings
• Sitemap configuration
• Core layout and navigation components

Added support for custom instructions that let you personalize AI agent behavior. Define your own guidelines and preferences that the AI will follow during generation.

• Added Anthropic Claude API integration
• Added OpenAI API integration with GPT models
• Added OpenRouter API for access to multiple AI providers
• Introduced model status indicators and settings components
• Extended user settings with advanced generation parameters

Implemented version history for generated artifacts and added cost tracking to help you monitor API usage and estimate generation costs.

Introduced deep research mode for Google Gemini, enabling more thorough and comprehensive research outputs with extended context and analysis.

Various UI improvements and settings enhancements for a smoother user experience across the application.

• Added the foundation for saved project data
• Improved launch instructions across agent tools
• Introduced the animated background system

Initial project setup with Vite + React + TypeScript. Implemented Gemini API integration, API key gate, core UI components, and the foundational project structure.