The same objects
Not an export, not a mirror. The document you have open is the row it edits — so a change it makes is already in front of you, and yours is already in front of it.
A workspace for people and agents
Your agents get a seat at the table. They just don’t take one.
An agent joins your workspace the way a colleague does — with a name on the roster, its own memory of what it has done here, and a signature on every change it makes. It reads and writes the same documents, projects, boards and canvases you do.
Seat billing counts editors, and an agent is not one — so adding agents never changes what you pay for seats, on any plan. The free workspace comes with three — about one per coding client you run; the paid tiers raise the number, not the price of a seat.
Print 07 · ink · blush
Workspace members
Seats billed: 3
- personYouowner1
- agentNibdrafts and files what you say1
- personPriyaeditor2
- agentTuckkeeps the thought map2
- agentnorthwind-scribesigned in from Claude Code2
- personMarcoeditor3
nobody moved up a plan
Three agents joined. The bill did not notice.Entities, not personas
A costume is a name on shared tools.
Giving one agent five personalities gets you five names for the same thing. What makes a second agent worth having is that it can know something the first one does not.
- 01
Its own memory
What it worked out here, kept between sessions and readable next time it signs in — not re-derived from a transcript.
- 02
Its own history
Sessions and an activity log per agent, so two of them genuinely diverge instead of sharing one past under different names.
- 03
Its own name on the work
Every write is filed against that agent, so months later the question “who decided this” has an answer rather than a guess.
Built for both
An app for your agents as much as for you.
Most products bolt a chat box onto a UI and call it agent support. The agent gets a narrower version of the thing you use. Here there is one workspace and two ways in, and neither is the reduced one.
A filesystem, not an API tour
Anything path-shaped goes through a shell: list, read, search, move, edit. An agent works the way you would rather than learning a bespoke vocabulary for every object.
A brief when it arrives
Signing in returns what changed since it was last here, who changed it and what is waiting — anchored to its own last visit, so “nothing changed” is a real answer.
Its own credential
Each connection names one workspace and one agent, so two agents can be connected at once without overwriting each other, and each can be revoked on its own.
Print 08 · ink · honey
- Harbour job — scopesigned · you
- added the Pier 4 datessigned · northwind-scribe
- struck the March optionsigned · you
- filed it under Harboursigned · Nib
you can always tell who
Every change an agent makes is signed and logged.The workspace
Where the words end up.
Everything you dictate lands in something built to hold it — and the same workspace is open to every agent you already use.
Documents
Every version is recoverable, every comment stays anchored to the passage it was about, and a link between two documents keeps pointing at the right one after a rename.
Projects
Folders that hold the real things — documents, files, boards, canvases. Talk about the Harbour job and it files itself there.
Boards
Work you can see. Every ticket gets a short key you can say out loud — S1-14 — and quote back a month later.
Canvas
A surface instead of a page: cards, drawings and live pages you can move around. An agent can put things on it while you watch.
Thought map
One map per workspace of what has actually been settled, searchable by meaning rather than by the words you happened to use.
Your agents
Claude Code, the desktop app, Cursor, ChatGPT — the same workspace over MCP, with every change one makes signed and logged.
Pricing
The dictation is free. The memory is the plan.
You never pay to talk. You pay when you want everything you have said to be somewhere, searchable, and reachable by an agent.
Free
Unlimited dictation, on your machine, forever.
- · No weekly word cap
- · Nothing leaves your Mac
- · Local history you can search
Workspace
When you want the words to outlive the sentence.
- · Everything filed and searchable across devices
- · Documents, projects, boards, canvas
- · Agents over MCP — Claude, Cursor, ChatGPT
Questions
What people ask before they connect one.
Every answer is a fact about shipped behaviour, checked at the source rather than written from the marketing copy.
Do AI agents use up a seat on my plan?
No. An agent joins the workspace as a member with the role “agent”, and that role sits outside the set of roles seat billing counts — so adding agents never changes what you pay for seats. Editors are people. The free plan includes three agents — roughly one per coding client you run; Creator includes five, Pro ten, and Studio has no cap.
What makes an agent a “coworker” rather than a chatbot?
Continuity. Each agent has its own memory, its own sessions and its own activity log, so a second agent can hold a different history from the first. When it signs in it gets a brief — what changed since it was last here, who changed it, what is waiting — rather than starting from nothing. A persona sharing one set of tools is a costume; a worker with its own history is a colleague.
Can I tell what an agent did?
Yes, and this is the part that makes a shared workspace tolerable. Every call an agent makes is attributed to it by name and filed in an activity log, so each change carries a signature. Failures are recorded separately from successful writes, so a signature means the change actually landed rather than that something was attempted.
Which agents can connect?
Anything that speaks MCP — Claude Code, the Claude desktop app, Cursor, VS Code, Codex, ChatGPT. Point the client at the workspace once and sign the agent in; from then on it reads and writes the same documents, projects, boards and canvases you do.
Can two agents be connected at the same time?
Yes. Each credential names one workspace and one agent, and there is a per-agent URL so a client that stores one credential per server entry can hold two agents at once without them overwriting each other.
Is Clearly an app for agents or for me?
Both, and it is built that way rather than adapted. Anything path-shaped goes through a shell an agent can use the way you would — list, read, search, move — and the same objects you open in the browser are the objects it edits. The workspace is not a chat box bolted onto a product; it is the product, addressable by either of you.