One place to see, review and make everything your agents do, and to shape it to how you work. Chat is where you ask. The rest of the workspace is where the work shows up: a version to approve, a plan to check, a calendar of what is coming, a library of what your agents can do. Every screen works on the same tasks and versions, so nothing you do in one is lost in another.
The screens below are grouped by what you are doing. Click any picture to see it full size.
Real screenshots of the app, except the chat view, which is a labelled illustration. Before capture, brands not covered by these docs, the owner’s name, personal items and money figures were hidden.
Ask for anything in plain words, on any screen or on your phone. The agent says which skill it is using, asks with a form when one short question decides the output, and hands back a card that opens the work.
Open it when you want something done, changed or explained.
Chat is where most work starts, but it is not where work is stored. A conversation sets goals, answers questions and redirects; the work itself lands in tasks and versions the conversation links to.
Private and shared. You have one private conversation per agent by default and can open more. What is said and remembered there is scoped to you and that agent. A shared conversation (a room) has explicit membership: add a partner or a client and they see the same thread and talk to the same agent. In a room the agent acts with the rights of whoever sent the current message, on the room’s brand only, and starts with no one’s private memory.
Watch the work as it runs. The conversation shows which workflow, which skill and which tool is running, nested as they call each other, so a five-minute silence is never a black box. Smart links turn anything the agent mentions into a card that keeps working when files are renamed, and browsing never steers: opening a task or a file does not change what the agent is doing.
Both work on the same tasks and versions. They differ in who is driving: in Review the agent made it and you judge it; in Make you create, and an agent can help.
Tasks
Every piece of work as a card: an inbox of what needs you, a kanban, a table, a weekly view and a checklist. Filter by business, agent, goal or stage.
Open it to see the whole plate, or to clear a batch of reviews.
Workbench
One task at a time: every version, its checks, comments and history. Approve, comment, rate, send it back in a sentence, or make a small edit yourself.
Open it when a version is ready for you.
Studio
Create in depth: one prompt bar for images, video and audio; a gallery of every generation; Elements (characters, places, props, styles) that keep a look consistent; and projects that assemble many takes into one piece.
Open it when a prompt in chat is not enough.
Mini-apps and productions
A few guided inputs over one skill, for when a form beats a prompt: a thumbnail maker, a character sheet. Productions are the larger version: a book into a short film, a product into a fifteen-second ad.
A channel run as a cycle, each with the same cockpit: overview, signals, ideas, the plan, patterns, data and playbook. Learn one and you know the rest.
Open it to approve what gets made next and to see what worked.
Schedules
Every routine your agents run on a clock: what fires, when, for which business, and where the result goes. Switch one on or off in a click.
A workspace you cannot change is someone else’s workspace. Two surfaces are built to be shaped by you.
A dashboard you compose
Add, move, resize and remove the widgets on Mission Control: tasks, metrics, goals, what needs you, quick actions, the daily review. Your layout is kept for you. Several dashboards as tabs — one for a business, one for your week — is Planned.
Click “Edit widgets” on Mission Control.
Mini-apps you build
Two mini-apps exist today. The direction is that you can turn any job you repeat into one: describe it, and the agent drafts a small form from the questions it actually asked you, over the skill that does the work. You approve it before it joins your menu. Building your own is Planned.
A mini-app holds no logic of its own — the skill stays the brain — which keeps it cheap to add and unable to drift.
Switching how involved you are should not mean switching systems or losing context. All five depths work on the same tasks, artifacts and versions.
Depth
You…
Where
Example
Describe and steer
State intent, answer questions, give feedback
Chat
“Three cover options, warmer than last time”
Compare and approve
Look at options with their checks; approve one
Workbench
Approve cover A; comment “warmer sky” on C
Adjust
Change a few things on an existing piece
Workbench, edit mode
Nudge the crop; tighten a paragraph
Run a mini-app
Do a frequent, well-defined job through a few inputs
Studio, a mini-app
Make five thumbnail options
Open a project
Create in depth: references, elements, rounds, assembly
Studio, a project
Illustrate a spread; cut a trailer
The version you commented on in review is the version a mini-app opens; what a Studio project produces lands on the task the conversation started; your rating attaches to the version you actually looked at, wherever you looked at it.
For builders — what holds the surfaces together
Skill, workflow, mini-app, project. A skill is the method for one job, written for an agent to read. A workflow is the sequence of skills, outputs and decision points that reaches an outcome. A mini-app is a declared form over one skill run, with no logic. A Studio project is the same idea at a larger scale: ordered steps, structured inputs, generation, on a task’s scene tree. A project is a task; one ingredient used in three scenes is one artifact tagged three times, never three copies.
One record underneath. Every surface reads and writes the same tasks, artifacts and versions, so a change made in chat, a mini-app or a project carries the same lineage.
Identity, not paths. Media and links are referenced by a stable identity, which is why a rename never breaks a card in chat.
Dashboards. Each widget is a saved row with its type, size, position and settings, per person; the board it belongs to is already a field, which is what dashboard tabs will use.
Related
How it worksYou ask, your agents do the work, you approve it.