Comment together, and let AI work in the background
Since rebuilding from the ground up for 1.0, we focused on how you work together and on making AI go deeper. Now you can comment anywhere on the canvas and in your documents, and AI can run in the background to take on bigger jobs.
💬 Comment Anywhere
On the canvas, you can drop a comment right onto a shell, box, or section. Hover over a block and the comment target appears; click and a pin anchors to its top-right corner. Click empty space and the comment stays pinned to that exact spot.
In docs mode, a comment rail floats on the right, Notion-style, with cards lined up next to each paragraph. You can also select text to comment on just that range.
The Comments tab in the right sidebar lists every thread at a glance, and clicking one jumps the view straight to it. Share a link to a specific comment and whoever opens it lands right on that thread.
Mention @Arky for help
Ask Arky from the same comment thread. Type a reply, press Tab, and Arky answers in chat without making you open a separate conversation.
🤖 A Floating Chat That Works in the Background
Press Ctrl twice on Windows or ⌘ twice on Mac, and an input pops up anywhere to ask the AI on the spot. Once you send, the input closes and a small pill in the top-right keeps working in the background. When the answer lands, the pill expands briefly to show a preview, and clicking it opens the full conversation.
Run several at once and keep working on other things. What's visible on screen and whatever you've selected are pulled into context automatically.
A richer conversation
- Attach text and code files, not just images
- Render Mermaid diagrams right inside the chat
- Control how hard the model thinks (reasoning effort)
- Project-scoped memories remember the context of each project
- On desktop you can connect your own API keys (BYOK), and there's a new Models tab in settings
- Use the latest models (Claude Opus 4.8, Gemini Flash 3.5)
🧠 Deeper Research and Skills
Deep research
Ask a question and the AI digs through the web and hands back a research card with sources. Citation chips and source favicons come attached, and clicking one jumps to the original. Jobs are now stored on the server, so a redeploy or a brief disconnect won't lose your results.
Skills
The AI can save procedures it uses often as Skills and refine them. Create one with skill-creator — before saving, it shows you which phrasings it expects to react to — and type /skills in chat to jump straight to the management tab. It can also pull in the skill it needs mid-conversation.
🔗 Embed Blocks
Drop in links from YouTube, Vimeo, tweets, or maps as embed blocks. When you paste a URL that can be embedded, the options popup suggests embedding it first.
📤 New Exports
You can now export to Word (.docx), and formatted blocks like tables and images come through intact. Exporting the canvas as a PNG image is in too.
🧭 A Smoother Canvas
- While creating an edge, click empty space to create a new node there
- Selections that scroll off screen are projected onto the edge so you know where they are
- Press
Mod-ato select all direct blocks under the current focus at once - Dragging is smoother, and shell and box heights now flex to fit their content
🔍 Library and Search
- Filter your library by tags, and narrow down fast with tag chips
- Import an entire Obsidian vault from a single zip
- Choose a personal workspace theme
- Duplicate a project
✨ Other Updates
- When you disconnect, a countdown shows the time until reconnect, and hovering turns it into a 'Retry now' button to reconnect instantly
- Click an image to open a preview with zoom and download controls
- Hover over a table to reveal buttons for adding rows, columns, and a next line
- Paste code copied from VS Code and it drops in as a code block
- Typing
=>or<=auto-converts to⇒and≤ - Polished sync stability, Korean (IME) input, collaboration cursors, and paste behavior throughout
