Changelog
New features, improvements and fixes. See what’s new in pen.dev.
Focus your conversations and expand your creative choices
New
- Expand desktop chat to fill the window, show or hide conversation history and return to the canvas when you are ready.
- Choose pen.dev, Recraft or QuiverAI for SVG generation from the model picker or Settings, with pen.dev free and Recraft and QuiverAI requiring a paid plan.
- Align text left, center, right or justified with keyboard shortcuts or command-palette actions, including while editing text.
- Calculate sizes and other properties directly in number fields with expressions such as 120*2 or (100-8)/2.
- Reload, navigate back and forward, and edit full browser URLs with keyboard shortcuts across desktop browsers, with a reload animation in presentation mode.
- Use pen.dev in French, Spanish, Brazilian Portuguese or Indonesian, with locally adapted example prompts in every non-English language.
- Choose Claude Fable 5.2 with Claude Code connected through a Claude subscription or Anthropic API key.
- Generate images with OpenAI Image 2.6 Flare and OpenAI Image 2.6 Sunburst on paid plans, with older OpenAI image choices moved to their matching 2.6 models.
- Enable Always steer messages to send new directions straight to a running agent that supports steering.
- Follow up to six running agents in activity bubbles while chat is hidden, and click a bubble to open its conversation.
Improved
- Manage parallel agents and up to five additional successive variations in Run settings, with Divide and conquer, Build separate versions and Layout, Style or Creative variation modes, within your plan's agent limits.
- Scan agent progress through expandable thinking and tool summaries, followed by shorter final messages.
- Agent requests start with less setup work, and Claude receives large instruction outputs correctly.
- Create more visually engaging boards with the whiteboard skill.
- Remove background and Vectorize image ask you to sign in when needed and continue after sign-in.
- Background removal, background replacement and image vectorization now require a paid pen.dev plan.
- Agents use their own tools to edit existing images instead of retrying pen.dev image generation.
- Presented browsers show their size toggle only when it can change the browser size.
- Start new conversations with the default model from pen.dev's available catalog when you have no saved model choice.
- Switch conversations from chat history; Ctrl+Tab conversation navigation has been removed.
- Agent chat no longer includes the dedicated TODO interface.
Fixed
- Screenshots and image exports reliably include your canvas content.
- Pinching or using Cmd/Ctrl+scroll over a canvas browser zooms the canvas immediately and leaves the page still.
- Double-click to present a browser reliably, including while its page is loading.
- Chat moves smoothly with the sidebar as it opens and closes.
- Settings, Goodies and sign-in dialogs keep focus when you use the command-palette shortcut.
Export vectors and turn notes into action
New
- Export selected layers or copy them as SVG, with shapes and text as vector paths, embedded images for unsupported effects and a ZIP of individual SVGs for multiple layers.
- Double-click a canvas browser to select and present it full screen.
- Let external agents connected through MCP save a document to a specified location with the new save tool.
- Choose desktop workflows for competitive research, recruiting profiles, sitemaps and travel plans, with canvas research alongside sources and guidance for responsive websites and staggered animations.
- Use pen.dev in Japanese.
- Use pen.dev in Czech, automatically selected when it matches your system language.
- Hold Alt/Option while picking a browser page element to inspect spacing between elements.
Improved
- Organize ideas with redesigned sticky notes in eight colors and a regular typeface, then send any note to an agent with Send to Agent; existing context and prompt notes become sticky notes.
- Ask agents to turn website designs into code and place the running site beside the design in canvas browsers so you can iterate on both.
- Use more of pen.dev in your language, including properties, browser controls, chat, sidebar and desktop menus.
- Find conversations in the chat sidebar's History list, with clearer Chat and New chat labels, better narrow layouts and tooltips for icon-only tabs.
- Find hosted-model conversations more easily with short titles generated from your first message.
- Use refreshed account, workspace and upgrade controls in both light and dark mode.
- Navigate refreshed Settings with improvements to dark and light themes.
- See which image models require a paid plan and open the upgrade dialog directly from their labels in Settings.
- Check agent connections at a glance with consistent green, grey and amber status indicators.
Fixed
- Keep hosted-model conversations going as screenshots accumulate, with older images omitted as needed and a clear prompt to start a new chat if the request remains too large.
- Share documents containing same-named files from different folders, with both files included.
- Access upgraded image models immediately as their upgrade labels clear without restarting.
- Document-tab close buttons fade smoothly on hover.
Find layers, follow agents and present live pages
New
- Find layers by name, text or ID, see matches on the canvas and step through results to select the right one, including unnamed layers by their translated type names.
- Choose GPT 6.1 Terra with a connected OpenAI ChatGPT agent.
- Copy any message you have sent to an agent.
- Use the Simplified Chinese interface beta, with translated Settings, commands, tools, menus and shortcuts.
- Cmd-click a link in a canvas browser to open its page in a new browser on the canvas.
- See how long an agent has been working and the total time when it finishes.
- Present a selected canvas browser full screen.
Improved
- Agent chat responds faster.
- Give agents context from your other open tabs and editors.
- Browse with a full-height browser panel while document tabs make room alongside it.
- Canvas browsers restore scroll positions and completed form fields more accurately.
- Pages appear sooner on sites with slow hidden frames.
- Selecting an empty browser keeps focus steady instead of jumping to its address field.
- Tooltips explain canvas browser action buttons.
- Scan agent work with tool-specific icons, success checkmarks and expandable details, plus clearer thinking labels and a steadier counter.
- Open File, Edit and View from a menu button beside the canvas tools on Windows and Linux.
- Review your usage in a refreshed Settings tab.
- Pan, zoom and edit more smoothly in large documents.
- Choose models from the keyboard while search keeps focus as you move the pointer.
Fixed
- Open external links from Home in your browser.
- Annual plans show the usage allowance period separately from the yearly billing period.
- Paste to replace preserves the layer's absolute position.
- Save As keeps fonts imported from websites into a draft during the same session.
- Image shaders from Goodies display images in the correct orientation.
Keep your documents and feedback together
New
- Work across documents in tabs within one desktop window, with Home alongside your designs.
- Discuss designs in pinned comment threads with images, move or resolve feedback, and ask the agent to apply selected threads after signing in.
- Find and run pen.dev actions by name from a searchable command palette with Cmd/Ctrl+P.
- Choose who can open a shared snapshot and whether viewers may comment or download the .pen file, then update those permissions without re-uploading after signing in.
- Use MiniMax M3.1 Flash through a connected provider that offers it.
Improved
- Large documents open faster and respond sooner, with a loading indicator shown until the canvas and fonts are ready.
- Browse recent files, drafts and design systems in separate Home sections.
- Switch themes, hide the interface and reach CLI guides or YouTube from the pen.dev menu, with Show UI available to bring the interface back.
- Start each new file with an empty canvas ready for your ideas.
- Choose from more models offered by connected providers such as OpenRouter, Together, OpenCode and Z.AI.
- Follow agent thinking with a live estimated token count.
- Find styles, shaders, design systems and scripts in a consistent alphabetical order in Goodies.
- Unsplash images clearly credit the photographer and source.
- Desktop agent conversations use more efficient storage for better performance.
- Eligible annual subscribers see the original monthly price alongside the early-bird price in the in-app upgrade dialog.
Fixed
- Website imports find fonts in chained stylesheets, skip unsupported font formats and save fonts, shaders and scripts in their own folders.
Explore shaders and keep large designs moving
New
- Explore 75 shader categories in Goodies, each with an example and four variants, then apply a fill or effect or insert the editable example; previews require internet access and respect reduced-motion settings.
- Use a layer's own content or its background in any shader image input, preview image inputs as thumbnails, and convert content-processing fills into effects that include children.
- Use Muse Wonder and Muse Spark through Meta with an API key or Muse subscription sign-in in the desktop app.
- Connect a SuperGrok, X Premium or Kimi Code subscription to the desktop agent by signing in.
- Press I to pick a color anywhere on screen and update the open picker or selected layer, or copy it when nothing is selected, where the system eyedropper is supported.
- Type @ in desktop chat to find files in your workspace folder and attach them to your message.
- Paste images and files directly into agent chat in the editor or Home.
Improved
- Navigate large documents faster when zoomed out while retaining full detail in exports.
- Get into large documents faster with improved opening performance.
- Activate a canvas browser with one click and identify it by its green tint.
- Layered background-reading shaders avoid unnecessary redraws when you zoom out below 50%.
- Follow external agents working through pen.dev's MCP server with visible cursors on the canvas.
- Understand each design step and terminal command with a one-line explanation in chat.
- Show or hide selected layers directly from the properties panel.
- Find your connections in refreshed Settings and profile controls, with MCP options grouped by how you use agents.
- Press Enter on a selected image to start cropping.
- Save dialogs suggest your draft's name when you choose a new location.
- Shadows and blurs render faster and more cleanly across zoom levels, effects over frames generate faster, and raster exports include background blur.
- Documents with many translucent shapes, icons and text redraw faster.
- Open and edit documents with fewer freezes while thumbnails are created in the background.
- Choose Import or Capture directly on a canvas browser, with Visible and Page capture options.
Fixed
- Undo and redo work more reliably with components and after adding, pasting or replacing content.
- Save your document reliably after an agent edit fails.
- Import affected web pages as usable canvas content.
- Document thumbnails show the saved version of your design.
- Double-clicking or Shift-double-clicking the canvas keeps focus on your canvas work.
- Codex chats preserve earlier steps and show messages once, with completed steps marked as finished.
- Launch pen.dev on Mac when the app is installed on an external drive.
- The Windows app stays responsive while agents work.
Connect your tools and fine-tune every detail
New
- Connect your own MCP servers to give every integrated desktop agent access to tools such as Linear, Notion and Mobbin, with browser sign-in for services that use OAuth and any service-specific requirements still applying.
- Replace selected placeholders with copies of your clipboard content.
- Rename editable local documents and drafts from the title bar or Dashboard, including the .pen file on disk.
- Choose GPT 6.1 Sol with a connected OpenAI ChatGPT agent, subject to the models available to your account.
- Choose Claude Haiku 5.5 in chat and canvas prompt notes with a connected Claude subscription or Anthropic API key, subject to your connection's available models.
- Find pen.dev on X, Discord support and the team's email address on the new Stay connected onboarding screen.
- Apply shaders as effects to any layer and its children, with live animation and multiple effects applied in order.
- Tune a selected shader's sliders, colors, images and vector inputs from a focused panel opened by clicking its name on the canvas.
Improved
- A refreshed properties panel brings consistent controls, new icons and clearer contrast in light and dark mode.
- Enter CSS colors such as red, rgb() or hsl(), and judge transparency with a split solid-and-checkerboard swatch.
- Show or hide each fill directly from its row in the properties panel.
- Adjust vertical and horizontal padding together, or switch to individual sides on flex-layout layers.
- Canvas text supports more writing systems, including Hindi, Bengali, Tamil, Armenian, Khmer and Amharic, plus math and symbols, without requiring a matching font choice.
- Switch the redesigned desktop browser panel between Laptop, Phone and Tablet sizes with one button.
- Edit large selections faster, with less redrawing while dragging thousands of layers.
- Use the Cursor agent on networks that block HTTP/2 by enabling Use HTTP/1.1.
- Generate images with Nano Banana 2.1 through the Nano Banana model option.
- Open web addresses and local files directly from links in agent messages and the prompt input.
- Find X, Discord support and the pen.dev website in a refined top-left menu.
- See what the recovered-backup badge means as soon as you hover over it.
- Configure Gemini from one Google provider card, with existing API keys carried over.
Fixed
- Text stays aligned with the canvas while you edit it.
- Undo and redo work after pasting, including content with component instances.
- Compose Chinese, Japanese and Korean text in place as the text box grows, and press Esc to cancel only the active composition.
- Properties show Mixed values correctly when selected layers have different settings.
- Clearing a number field preserves its intended empty state, and arrow-key editing works correctly.
- Removing a variable preserves the property value, and variable pickers toggle correctly.
- Padding controls apply to the correct sides, and rotation controls follow the expected direction.
- Undo restores Hug and Fill sizing correctly, and group sizes display accurately.
- Property controls respect which edits the selected layer supports.
- Component headers display correctly in light mode.
- Animated shaders play on the current presentation slide, including after advancing to the next slide.
- Documents with shaders that read texture sizes on large layers stay open reliably.
- Removing a Cursor API key shows Not connected immediately.
- Pinch and Ctrl-scroll zoom the canvas over embedded frames inside browser nodes.
- The welcome video plays on Windows, and its margins blend into the desktop app background.
- Text selection stays accurate when you drag beyond the text box.
- Text line height follows the main font.
- Use the Cursor agent in the Intel Mac app.
Crop images, steer prompts and present at actual size
New
- Get the crop you want by moving an image and its frame independently, with Stretch, Fill, Fit and Reset crop controls.
- Choose an image generation model directly from the agent model picker. In the web editor, sign in with access to the provider.
- Keep ideas moving with editable, removable queued prompts, and use Steer to send the next prompt immediately to a running Claude, Gemini or Pi agent.
- Preview slides at their actual size with 100% zoom, or fit them to the screen in presentation mode.
- Ask your agent to click, type, scroll, navigate and read values on pages in the desktop app's built-in browser.
- Turn research, comparisons and plans into a canvas whiteboard, with live source pages, findings alongside them, a separate result frame and open questions for you to decide.
- Open pop-up windows from the built-in browser after reviewing and allowing the request.
- Work with Claude Sonnet 5.5 through a connected Claude account or Anthropic API key, with saved Sonnet 4.5 and 4.6 choices moving to Sonnet 5.5.
- Choose Grok 4.7 through a connected xAI, Open Router, OpenCode Go or eligible GitHub Copilot account. In the web editor, use an API key for a browser-compatible provider that offers the model.
- Explore new pen.dev pages explaining workflows for Figma, AI design, slides, dashboards, mobile, web and web apps.
- Open the web editor directly at app.pen.dev.
- Choose GPT 6.1 Sol from supported agent providers.
Improved
- Explore 89 design styles, up from 53, with refreshed previews to help you choose a direction.
- Keep your bearings with animated canvas movement when zooming, entering or leaving presentations, and bringing layers into view.
- The canvas brings an off-screen browser address bar into view when you focus it.
- Set an exact width or height directly in the properties panel, even when that dimension uses Fill or Hug.
- HTML exports preserve fill blend modes without turning those fills into images.
- Agent panel guidance makes saving a design and generating code separate, clear steps.
- Documents with many shaders stay responsive as they open, with shaders appearing sooner.
- Scrolling and editing require less work to update the interface.
- Long agent conversations take less work to display by rendering the messages in view.
- A smaller pen.dev logo gives the title bar a more compact appearance while preserving its proportions and alignment.
- The web editor reuses its compiled engine when you return.
- Canvas rendering avoids a second redraw when the layout also changes.
Fixed
- Screenshots and exports include newly added layers, including previews captured by your agent.
- Watch the complete design script build up as your agent writes it in chat.
- Pasted layers and images land in the middle of the screen.
- Entering and leaving presentations keeps keyboard focus out of the agent prompt, so the presentation shortcut works as expected.
- New opens your existing empty conversation before creating another one.
- Audio and video pause when a canvas browser moves into the background.
- Importing app-style pages such as YouTube brings their content onto the canvas.
- Selecting a browser with a slight mouse movement leaves the canvas ready for your next action.
- Live shares finish syncing after bursts of edits and send pending updates when an agent finishes (for signed-in owners editing their own browser documents that automatically sync to a share URL).
- Agent execute calls accept unused optional parameters sent with default values.
Add inner shadows and find your image model
Improved
- Pending invitation limits now apply per inviter across organizations and workspaces, including invitations sent simultaneously; existing limits within each organization and workspace still apply.
- The model picker lists pen.dev models before other providers.
- Agent help shows browser-specific HTML export instructions and the correct attachment control.
Fixed
- Document validation errors use the pen.dev name.
- Fill removal and variable controls stay visible alongside long variable names.
- The Variables panel shows a variable’s new name as soon as you rename it.
- Your requested sidebar tab returns when an agent becomes available.
New
- Add inner shadows to objects from Properties > Effects > Inner Shadow.
- Choose OpenAI Image 2.5 Flare or Sunburst as your default image-generation provider in Settings > Import & Generation (requires sign-in and provider availability).
Scale groups, edit images and import websites
New
- Scale editable groups together with their text, strokes, corner radii and effects using the Scale tool (K).
- Paste a website URL into the web editor to bring its layout onto the canvas as editable layers.
- Remove an image background or turn a bitmap into editable vector paths.
- Choose how far the arrow keys move a selection in Settings.
Improved
- Tools and editor panels now float over the canvas with the pen.dev identity.
- Sidebar buttons and provider-status banners use consistent sizing and spacing.
- The pen.dev website has updated layouts, cards and copy.
- Settings and Agents controls are now in the title bar, and the sidebar opens wider by default.
Fixed
- The Agent tab label displays in full, and switching tabs animates consistently.
- Floating panels, including Variables, resize diagonally.
Run CLI sessions side by side
Improved
- Run multiple CLI sessions at once without one session taking over another’s connection.
Capture more of a webpage
Improved
- Web imports now pick up lazy-loaded content as the page scrolls.
Fixed
- Copying a webpage with the Chrome extension no longer gets stuck on “Copying…”.
Earlier releases
- Copy a webpage with the Chrome extension and paste it into your design to edit the text, images, and layout.
- Flip a selection or rotate it by 90 degrees.
- Connect a custom model provider and set its context window.
- Open exported HTML in your browser as soon as the export finishes.
- Capture a screenshot in the built-in browser to keep a visual reference beside your design.
- Jump from an instance to its component, or copy a selection as JSON, from the canvas context menu.
- Edit gradients directly on the canvas.
- Imported SVGs preserve more gradients, transforms and class-based styles.
- Point the Claude agent at a custom executable in Settings.
- Export a design as code from the editor, including when working inside VS Code or Cursor.
- Include the canvas background when exporting an image.
- Hold Alt while dragging to duplicate a selection.
- Reopen a recent file from the file picker.
- Attach a PDF to an agent conversation so the agent can work from your brief.
- Run design prompts from the CLI with Codex or Gemini.
- Copy a selection as a PNG with a keyboard shortcut.
- Continue copying and pasting on the canvas after working in agent chat.
- Draw polygons directly on the canvas with the polygon tool.
- Export a snapshot to package a design with its assets.
- Explore 32 more prebuilt design styles.
- Sign up for an account from the CLI.
- The pixel grid fades in as you zoom closer.
- Work through a design in an interactive CLI session and export individual nodes from the terminal.
- Export larger images with smoother edges.
- Adjust corner radii independently on the canvas.
- Resize the sidebar or minimize chat to give your design more room.
- Open a .fig file from the Figma import dialog to bring the design onto your canvas.
- Split design work across agents so several parts of a canvas can take shape at once.
- Return to a saved agent conversation.
- Set the workspace folder for a .pen file.
- Connect GitHub Copilot to pen.dev through MCP.
- Image imports now load correctly in the desktop app.
- Browse visual previews when choosing a design style.
- Blend colors across a shape with mesh gradients.
- Agents can create and update multiple layers in a single batch, keeping related edits together.
- Gradients from Figma import more faithfully.
- Ask an agent to find a stock image or generate one while building a layout.
- Copying a component preserves overrides on nested layers.
- Open .pen files directly in the standalone desktop app.
- Copy and paste layers from the layer list using keyboard shortcuts.
- Shift-select layers in the layer list and drag them into order.
- Undo and redo edits to components and instances.