Updated June 10, 2026
Not another markdown editor
You already have six places to write. Here's why Drafty added a seventh anyway.
You don't need another place to write markdown. You have Notion, Apple Notes, Obsidian, VS Code, and a few more you've half-abandoned. Writing was never the problem.
The problem is that none of those pages are shared with Claude. Your drafts live where Claude can't see them; Claude's drafts land where you can't type. So the two of you collaborate by copy-paste — out of chat into your editor, out of your editor into chat — and every paste forks the document into two copies, neither quite current.
A Drafty canvas is the page Claude publishes to — a real link, with comments, history, and sharing already attached. Edit mode makes it a page you can write in too. Which means this was never really an editor we were adding. It's the missing half of a shared surface.
Introducing edit mode
Markdown canvases are now directly editable by their owner — the Edit pill in the header on desktop, the pencil on your phone. Hit it and the rendered page you were reading becomes the page you're writing: same layout, same typography, with a caret where you click. There's no save button. Changes save as you type, a quiet Saved confirms it, and closing the tab mid-sentence is safe.
Two moments are why it exists.
Draft it, then hand it over
Some documents start with you. A half-formed plan, five rough bullets, the skeleton of a spec — not worth a polished write-up yet, but worth getting out of your head.
Start a New canvas from your canvases and sketch it — the canvas names itself after your first heading. Then turn to the Claude you already work with and say: turn this into a real plan.
That's the whole handoff. Your draft was already in Claude's world — same link, no paste, no upload, no "let me share a doc with you." The agent pulls the canvas, does the thinking, and pushes the fleshed-out version to the same link, with your rough cut preserved as version one.
Fix a word in what Claude made
The other direction: Claude ships a plan, and it's right except for one thing. A number. A date. A heading that says this week when it should say next week.
That fix used to cost a prompt — explain the change, wait for the model, get a fresh version — or a round-trip through copy-paste, after which the canvas and your copy disagree. Now it costs what it should:
Click the word, type the word. No tokens, no waiting, no second copy. The canvas stays the single source of truth, and your fix is recorded like any other change.
Saving is continuous, versions are not
The obvious worry with no save button is history: does every keystroke become a version? No — saving and versioning run at different speeds, on purpose.
Keystrokes flow into the live canvas continuously. A version is cut when your editing session ends — you pause for a couple of minutes, hit Done, or leave the tab. One sitting, one entry in History: edited on canvas, with your name on it. It's the same History that Claude's pushes write to, with the same restore — a bad edit is undone exactly like a bad push.
You and Claude won't collide
The sharp edge of a genuinely shared page: your agent ships versions of the same document. What happens when a push arrives mid-sentence?
It doesn't. While you're typing, the canvas is held — a push from the CLI is refused with a clear "someone's editing, retry shortly" instead of silently overwriting your work. The hold follows your activity: it releases a couple of minutes after you stop, or immediately on Done. Claude retries, the push lands, and your edit is already safe in History.
And if a change does land underneath you — say, you're editing the same canvas in two windows — the editor notices, stops, and asks which version wins, with your text intact either way. Nothing in this system deletes anything: every save path ends in append-only history, so the worst possible collision costs a click, not a paragraph.
What stays Claude's
HTML canvases don't get the toggle — an interactive artifact isn't a text document, and pretending it is would mangle it. The comment loop stays the edit path there: pin a note to the element, let the agent do the surgery.
The same honesty applies to the rare markdown canvas with raw HTML blocks embedded in it. The editor round-trips everything markdown can say — headings, lists, tables, task lists, images, code — but an HTML island would be flattened on the first save, so those canvases keep their pencil hidden rather than quietly destroying your formatting.
It works where you read
Most canvas reading happens on a phone, so editing had to work there too — not as a desktop feature that technically renders. Canvases use more of the screen on mobile, the header keeps just three actions (edit, share, everything else), and the editor is built for touch: tap to place the caret, type, watch Saved tick, hit Done.
Get started
Start a canvas from your canvases and hand your next rough draft to Claude without a single paste — or open anything Claude has already made for you and fix the word that's been bugging you. Either way, you're both writing on the same page now.
Related help
- Can I edit a Drafty canvas directly in the browser?Markdown canvases you own on Drafty have an Edit toggle for editing right in the browser. Changes save as you type, each editing session becomes a version, and Claude's pushes wait while you write.
- Drafty canvas version history: push, restore, and compareEvery push saves a new canvas version. Preview and restore earlier ones from the CLI — the link and comments stay put.