A graph you can see
Nodes connected by edges. Lay them out, wire up what's related, and the structure of your thinking becomes something you can actually look at.
A canvas where humans and agents share the same surface.
Build it together. Hand it off. Step back through every change. ActionGraph is a real-time collaborative diagram where every action — yours or an agent's — is part of one shared graph.
How it works
A graph of live nodes — that you and your agents edit together.
ActionGraph is a shared canvas where your work is a graph: things you're working with, and the relationships between them. Here's the shape of it.
Group, nest & embed
Group nodes in containers. Turn any container into a nested diagram — like opening a sub-directory. Embed diagrams from other workspaces right onto the canvas.
Every node is live
Not just boxes. Shapes, plus interactive components on the canvas: YouTube, websites, images, video, post-its, markdown, spreadsheets, and PDFs.
The canvas is also text
View and edit any diagram as XML or JSON, save it, and it renders right back on the canvas — the same round-trip you know from HTML.
Humans & agents, together
Claude Code, Cursor, opencode, Pi — or your own harness — connects, authenticates, and works the canvas alongside you. Same controls, no separate AI mode.
Shared in real time
Share a canvas and everyone joins the same live surface. Edits, agent actions, and new content show up for everyone instantly — one graph, always in sync.
What you can do with it
Built for the kind of work that's part planning, part research, part execution.
Investigate any video, ticket, or page.
Drop a YouTube link. Get transcript + AI summary + extracted entities, wired together. Have an agent dig deeper.
Write a plan with the sources wired in.
A plan node connected by edges to the meetings, tickets, and research that informed it. Click any edge to see why.
A canvas that grows with what you're working on.
Real-time collaborative. Synced for everyone. Survives sessions. Connect this week's work to last quarter's.
The wedge
Most "AI canvas" tools bolt an assistant onto the side. We didn't.
Agents work the canvas through the same controls you do. There's no separate AI mode. When an agent moves a node, it's the same move you'd make — you see it instantly, you can undo it, and it lives in the same shared history.
That means you can build something together, hand it off, and step back through every change either of you made.
You move a node
- → it moves on the canvas
- → everyone sees it instantly
- → you can undo it
- → it's in the history
An agent moves a node
- → it moves on the canvas
- → everyone sees it instantly
- → you can undo it
- → it's in the history
Free beta · invite-only
Want access? Tell us what you'd build.
We're inviting a small group of testers. If you've got a real workflow you want to put on the canvas — research, planning, agent harnesses — we want to hear about it.