Hold ⌘ to select layers inside a frame, drag several layers together, and see guides while dragging
Hold ⌘ (Ctrl on Windows) and drag on a frame to draw a selection box inside it and select the layers it touches, without moving the frame. Starting the drag on a layer selects within the level that layer belongs to. Hold Shift+⌘ to add to the current selection. A ⌘-click without dragging still selects the innermost layer directly.
After selecting several layers in a free-layout frame, dragging one of them moves the others along with it, and ⌘Z undoes it in one step. Layers dragged together stay in their frame; they aren't moved into another frame or out onto the canvas.
Dragging layers in a free-layout frame, one or several, now shows guides just like dragging objects on the canvas, snapping to the edges and center lines of the frame, nearby layers, and other objects on the canvas. Snapping can be turned off in canvas settings.
October 27, 2026v0.265.2
Fix: the breathing glow between pieces sometimes ended early while an agent drew a page
Fixed: since 0.265.0, the breathing glow on the page frame after an agent finished a piece sometimes disappeared after a few seconds instead of lasting until the next piece, for example after switching to another page and back in the page list, or when the collaboration connection briefly refreshed. The glow now carries on in these cases until the next piece starts, the agent starts a screenshot, or at most 45 seconds.
October 26, 2026v0.265.1
Fix: dragging a layer inside a frame no longer targets the outer frame
When you drag a layer inside a free-layout frame, it now stays in that frame and follows the pointer as long as the pointer is inside the frame. The outer frame no longer shows a highlight and insertion line, and releasing no longer moves the layer into the outer frame. To move it out, drag the pointer outside the frame.
Dragging a nested layer no longer shows a dashed hover outline on the outermost frame.
October 26, 2026v0.265.0
The canvas animation no longer stops between pieces while an agent draws a page
When an agent draws a page piece by piece, the page frame on the canvas now keeps a soft breathing glow between pieces, and the status next to the agent's cursor stays visible, instead of going dark as if the agent had finished. The usual animation returns as soon as the next piece starts.
When the agent takes a screenshot to check its work, the glow fades right away. Without a screenshot it lasts at most 45 more seconds, fading over the last 10. It disappears immediately if the agent disconnects or leaves the page.
This applies both to agents connected to M2 and to the built-in agent in the M2 panel.
October 26, 2026v0.264.0
Agents are now guided to draw a page frame first, then fill it in piece by piece
M2's instructions to agents for drawing a page have changed: place the page frame on the canvas first, then add the navigation, title, filter bar, table, pagination and so on one piece at a time, instead of writing the whole page in one go. Drawn this way, content shows up on the canvas sooner and keeps arriving, rather than after minutes of waiting. How an agent actually draws is still up to the agent.
The instructions now also say to check the page with one screenshot once it is done, instead of after every piece.
When adding content into the frame, agents now get the IDs of the frame's areas directly, so content is less likely to end up in the wrong place.
Agents that are already connected need to reconnect to M2's MCP to see the new instructions.
October 26, 2026v0.263.0
Fix: Copy as PNG shifted and cropped content; pasted images shrank
Fixed: selecting several frames with shadows and using Copy as PNG shifted everything down and to the right and cut off the right and bottom edges. The image is now complete and matches the canvas, including rotated layers and layers inside frames.
Copy as PNG now keeps shadows and other effects drawn outside a layer, and trims the extra transparent space around them. Layers without such effects copy at exactly the size of the selection.
Pasted and dropped images, and images dragged from Resources onto the canvas, are now placed at their original size instead of being shrunk to 800 px wide. Copying as PNG from M2 and pasting back gives the same size as the original layers.
Hidden layers are left out of Copy as PNG. Copying only hidden layers shows a "copy failed" message.
October 26, 2026v0.261.0
Agents know which frame and text properties they can use
Agents connected to M2 (the built-in Agent and your own) can now see which properties and values frames, text, images and icons support before they start drawing — for example per-side padding, per-side border widths and strikethrough text — instead of guessing or retrying.
Fixed: when an Agent added a frame with per-side padding, row or column gaps, or min / max sizes, or gave text an alignment, a line limit, italics or strikethrough, it was wrongly told the property was unknown.
Fixed: when an Agent changed an existing image's fit mode, aspect ratio, fallback image or loading placeholder, the change was dropped.
When an Agent writes a property that has no effect (such as textDecoration or padding on text), the warning now tells it the right property to use.
When an Agent writes with canvas_add, canvas_manage_children, canvas_modify_node or canvas_batch_update_children, wrong values for text alignment, strikethrough, italics, line limit, width and height, and for frame corner radii, min / max sizes, opacity and clipping (for example strikethrough written as "line-through", a width of "100%", or an opacity of 50) are now rejected with the list of valid values, instead of being saved without effect.
Agents that are already connected need to reconnect to M2's MCP server to see the new property list.
October 25, 2026v0.260.4
Fix: undo after background page cleanup
Fixed: if a page was cleaned up in the background (automatic slimming, edit-history cleanup) while you were away, pressing undo after coming back could undo only half of an action and leave two copies of the same layer. Now, in that case, actions from before you left can no longer be undone (the undo button is disabled), so undo no longer creates duplicate layers this way.
While a page is reconnecting and catching up with the server (including while you are offline), you can't undo actions from before you left; most things you do in the meantime can still be undone as usual, but deleting layers that existed before you left, or dragging them into or out of frames, can only be undone once syncing finishes. Once syncing finishes, if the page wasn't cleaned up, those steps and actions from before you left can be undone again; if it was, they can no longer be undone.
October 25, 2026v0.260.3
Updated Free plan details on the pricing page
The Free plan on the pricing page now lists 500 MCP calls a week, 3 projects with 3 pages each, 50 MB of assets, 2 editors and unlimited viewers.
The FAQ now says viewers are unlimited on every plan.
October 24, 2026v0.260.2
Fix: Chinese and Japanese pages switching to English after sign-in
Fixed: if you were signed in and your account language was English, the Chinese and Japanese pages of m2ui.ai switched to English a couple of seconds after loading. They now stay in the language of the page you opened, and the app still follows your account language once you open your workspace.
October 24, 2026v0.260.1
Fix: Docs pages redirected to the overview
Fixed: opening the user guide or the MCP setup guide directly (m2ui.ai/docs/guide, m2ui.ai/docs/mcp) sent you to the Docs overview instead. Both pages now open as expected.
October 24, 2026v0.260.0
m2ui.ai in Japanese
The website is now available in Japanese at m2ui.ai/ja: home, pricing, docs, the user guide and the MCP setup guide. Pick 日本語 from the language menu in the footer.
The Privacy Policy and Terms of Service are shown in English on the Japanese site, and release notes on What's new stay in English.
The M2 logo in the top bar now takes you back to the home page in your current language.
October 24, 2026v0.259.0
New m2ui.ai website
New home page: see how your team and AI agents design together on one canvas, with live examples of multiplayer editing, the built-in agent, design systems and connecting your own agent.
New Pricing page with Free, Lite and Pro plans, plus answers to common questions about seats, credits and billing.
New Docs: a user guide that walks through the editor, the built-in agent, design systems, sharing and exporting, and the MCP setup guide now lives at Docs → Connect over MCP (old /mcp links still work).
New What's new page with the full product changelog, grouped by month.
Updated Privacy Policy and Terms of Service, and a new site footer with company and contact details.
October 24, 2026v0.258.7
Internal: groundwork for an undo fix
No visible changes. Groundwork for an undo fix in the next release.
October 24, 2026v0.258.6
Internal: position guard for full node replacement
No visible changes. Fixed an internal safeguard so that a full node replacement checks a vector shape's actual position and never moves a repositioned vector shape back to its old place.