Introducing Docs

Today we’re launching Docs, a collaborative markdown editor built around one observation: the documents that matter most right now are the ones humans and AI agents work on together — and almost no editor treats both as first-class users.

Google Docs is wonderful for humans and nearly opaque to agents. A markdown file in a git repo is wonderful for agents and lonely for humans. Docs sits in the middle: a live multiplayer editor in the browser, and a plain REST API underneath, both operating on the same document at the same time.

One link is the whole product

Create a document at docs.aicomputercompany.com — no signup, no workspace setup, no invitation flow. You get a URL like:

https://docs.aicomputercompany.com/<doc_id>

That URL is the entire permission model. Send it to a teammate and they can edit. Paste it to your AI agent and it can edit too, over REST, with no API key. The document id is a long, unguessable token; possession of the link is the authorization.

We call this a capability URL, and we chose it deliberately. Every additional step between “I have a thought” and “we’re editing the same document” is friction, and friction is where collaboration dies. For this beta we picked the most radical point on the spectrum: zero accounts, zero keys, zero gates. (The honest trade-off: anyone with the link can edit. Our FAQ spells this out — don’t put secrets in a beta product.)

Multiplayer that survives robots

Docs is built on Yjs, the same family of CRDT technology behind the best collaborative editors. Live cursors, presence, instant convergence — the things you expect from multiplayer editing.

The part you might not expect: it holds up when one of the collaborators is a machine. When an agent edits a document through the API, the server applies the change as a minimal diff against the live shared document, not a wholesale replacement. Your cursor doesn’t jump. Your selection doesn’t vanish. You can literally keep typing a sentence while an agent restructures the section above it.

This sounds like a small thing. It is the entire difference between an editor agents are allowed to touch and an editor you’d actually let them touch while you’re working.

More than text: review, suggestions, views

Editing is the floor, not the ceiling. Docs ships with a review layer designed for AI collaborators:

• Annotations with verdicts — agents anchor notes to exact passages and tag them valid, question, or conflict, so a reviewed document has a visible health map.
• Suggestions — tracked changes a human can accept or dismiss, instead of silent rewrites.
• Comments and threads — discussions between any mix of humans and agents, attributed and resolvable.
• History and checkpoints — named versions and rollback, so handing a document to an agent is never a one-way door.
• Dynamic views — agents can push entire custom HTML interfaces, rendered in sandboxed iframes beside the document. More on this in its own post.
• Clean export — one call strips all review markup and returns publish-ready markdown.

Built to be discovered by agents

We want an agent with no prior knowledge of Docs to be able to use it. The product serves llm.txt and a full agent manual at llms-full.txt, and our home page has a copy-paste prompt block that teaches any AI the complete lifecycle in about forty lines: create, read with line numbers, edit with str_replace semantics, review, checkpoint.

If your agent can make HTTP requests, it can use Docs. There’s no SDK because there’s nothing an SDK would add.

What’s next

This is a Phase-1 beta, and we’re saying so plainly: beta-grade infrastructure, capability links instead of accounts, free while we build. Teams, permissions, and pricing come later. What we wanted to ship first is the thing we couldn’t find anywhere else — a document that’s equally alive in a human’s browser and an agent’s tool calls.

Try it: create a document, paste the link to your AI, and ask it for a review. It takes two minutes, and the two minutes are the pitch.