How it works
One document.
Every mind in the room.
Stateroom is a native shared document where people and agents work side by side. There is no chat window and nothing to send. To ask for help, you simply write a mention — and the answer takes shape right where you asked for it.
01 / The idea
The work lives in one place — and so does the thinking.
The session is the document
Plans, drafts, decisions and evidence accumulate in a single living page. Nothing important is scattered across threads you have to reconstruct later.
Agents are colleagues, not chatbots
An agent appears in the margin like any other collaborator, reads what it is given, and writes into the page beside yours — while you keep typing.
Trust you can see
Everything written carries its author. What the machine proposes stays visibly distinct until a person decides it is settled — and only a person can make that call.
02 / A summon, moment by moment
You write two words. The room does the rest.
- You type @agent where the question lives.
A mention is not a message bolted to the side — it is part of the sentence you were already writing.
- A colleague joins the document.
The agent takes a cursor of its own, exactly like a person would. You see it thinking; it sees your latest words.
- It reads only what it should.
The region around your mention becomes its scope. Your context stays deliberate, never accidental.
- Its answer arrives as real text.
Not a bubble — paragraphs in the page itself, which you can select, comment on, or rewrite even as they appear.
- Its work stays separate from its words.
Everything the agent did along the way is folded neatly beneath its reply: visible when you want the receipts, invisible when you want the prose.
- You decide what is true.
Certify a passage and it carries a quiet mark of trust. Edit it later, and every copy in the room honestly steps back to untrusted until re-earned.
03 / The floor beneath it
Six quiet systems, one calm surface.
None of them ask for your attention. Each exists so that the page in front of you can be trusted, continued, corrected, and shared without ceremony.
Who wrote what, what was trusted when, and how every correction points at what it fixed.
CollaborationColleagues that write↗Agents with their own cursors, writing beside humans instead of talking at them.
CompatibilityYour agents, invited↗Existing agent tools join the room through an open standard — no rebuild required.
IntelligenceAny model, one door↗The thinking behind the cursor is swappable. Your workflow never notices.
PresenceTogether in real time↗Everyone on the same page as it changes — offline edits reconcile without drama.
CraftA canvas for thought↗One continuous surface, drawn fresh every frame. No panels, no chrome, no noise.
04 / What we promise
A few things we refuse to do.
Nothing pretends. If a mark of trust cannot be verified — connection lost, history unclear — it degrades honestly rather than reassuring you falsely.
You hold the pen. Machines may propose all day; only a person settles anything. An attempt by an agent to bless its own work is refused, politely, and remembered.
Corrections add; they don't erase. Fixing a mistake leaves the mistake findable. History is a chain, not a chalkboard.
Your undo touches only your work. Rewinding your edits never rolls back someone else's — human or machine.