The ask lands as edits to the frames in view, never as a fresh file beside them.
The AI design agent that edits your file, not a copy of it
Ask for a change and that change lands on your canvas as edited frames, every element still a layer. Your fixes hold and your undo history stays yours.
The agent that edits your file instead of chatting about it.
The shared live canvas: frames the agent wrote sitting beside frames people finished by hand. Click in and drag either kind.
An AI design agent should edit, not answer
Most AI design tools are chat rooms with a render button. You describe, they generate, you spot the one wrong thing, and the next generation loses the three things that were right. Shyne's agent skips the conversation about the file and works in the file.
It reads the frames already on your canvas, then writes real HTML frames back onto it. Ask it to fix a heading and it fixes that heading. The rest of the canvas, including everything you touched by hand, stays exactly where you left it.
It writes layers, so you can take over anytime
Every frame the agent writes is ordinary HTML, and the editor promotes each meaningful element into a layer you can select, drag and restyle. When it gets a frame most of the way there, you finish it by hand instead of describing the last ten percent in prose.
Its changes also land in the project history like anyone else's, so undo stays safe: rolling back your last edit never swallows its work, and its pass reverts without touching yours.
Its edits sit in your history like any teammate's.
Four verbs it handles in place
Restructure
Reorder frames, cut a section, merge two pages. The document changes; no copy appears.
Restyle
Swap type, air out spacing, apply the brand tokens. Content stays put while style moves.
Extend
Ask for a closing frame that echoes the cover and get one more frame in the same system.
Repair
Point at the frame that overflows and it reads that frame, then fixes exactly that frame.
A margin note becomes the edit itself; the paragraph is rewritten where it stands.
Every AI works on the same design agent canvas.
Bring the agent you already use. It reads the brief, sees the work and writes the next editable frame — no exported file, no paste, no drift.
Restyle the whole deck to the new brand and fix the overflow on frame three. build the complete working system.
Twelve frames edited, layers intact is now composed on the same live canvas.
Restyle the whole deck to the new brand and fix the overflow on frame three. create three useful variants.
Twelve frames edited, layers intact is now composed on the same live canvas.
Restyle the whole deck to the new brand and fix the overflow on frame three. compose the visual direction.
Twelve frames edited, layers intact is now composed on the same live canvas.
Restyle the whole deck to the new brand and fix the overflow on frame three. turn the approved frame into code.
Twelve frames edited, layers intact is now composed on the same live canvas.
Restyle the whole deck to the new brand and fix the overflow on frame three. prepare the publishing pass.
Twelve frames edited, layers intact is now composed on the same live canvas.
Restyle the whole deck to the new brand and fix the overflow on frame three. run your own workflow on the canvas.
Twelve frames edited, layers intact is now composed on the same live canvas.
One shared canvas
Every design agent frame stays visible to people and agents at the same time.
Always on brand
Your colors, type and reusable assets stay attached to every design agent variation.
Editable output
The result lands as real frames and objects, ready for the next human decision.
Media in context
Generate or place imagery where the composition needs it, without leaving the working surface.
Open to every agent
Claude, ChatGPT, Gemini, Cursor, Grok and your own agent work through the same MCP surface.
Ready to ship
Move from the approved canvas to the useful file, link or published format without rebuilding it.
Direct it like a colleague with the file open
The agent holds the same canvas you do, so a request comes back as changed frames you can inspect, nudge or undo on the spot. Nothing arrives as an attachment.
- Move the case study above the pricing sectionThe frames trade places on the canvas in front of you; styles ride along.
- Make five more frames in this visual systemYour strongest frame becomes the brief; five siblings appear inside its grid and palette.
- Swap every headline to the condensed faceOne pass over the type layers; your layout and copy edits are untouched.
Ask for a tighter deck and the same ten frames change; nothing is rebuilt behind your back.
Bring your own agent to the same canvas
Shyne speaks MCP, so Claude, ChatGPT, Gemini, Cursor, Grok or an agent you built can open the project, read the frames and write edits into place. Whoever asks, the canvas stays the single copy of the work.
One canvas, any agent: Claude, ChatGPT, Gemini, Cursor or Grok.
Questions
What is an AI design agent?
Software you direct with plain requests that edits a design file the way a person would. In Shyne it reads the frames on your canvas, writes real HTML back, and every element it touches stays an editable layer.
Will the agent overwrite my manual edits?
No. It edits in place rather than regenerating, so a request about one frame changes that frame and leaves the rest alone. Undo is safe too: its changes and yours roll back independently.
Which AI models work with Shyne?
The built-in agent works out of the box, and the canvas speaks MCP, so Claude, ChatGPT, Gemini, Cursor, Grok or your own agent can read and write the same project.
Can the agent match work I already designed?
Yes. Because your frames are the document it reads, you can build one by hand and ask for more in that style. It continues the system on the canvas instead of guessing from a text description.
Put the agent to work
The agent that edits your file instead of chatting about it.
Open Shyne →
