One board for your team and your agents
Every other AI agent answers “how do I get this task done?” — in one person’s editor, terminal or chat window. This one answers where the work lives: a shared board per project, your colleagues and as many configured agents as you need working the same cards.
The real product, running on example data. Avatars are the people on a card; the stripe colour is the agent that ran it.
The board is the shared surface
Invite colleagues into a project and you all open the same board. Cards group by area of work, each card carries its conversation with the agent, and anyone with write access can add to that conversation. The people who worked an item are shown on it, and the agent that ran it tints it, so a glance tells you what is in flight and what produced it.
Filter the board down to one person when you want their view, or leave it on everyone when you want the company's. Work that should not be public still has a home: every member with write access gets a private card only they can see.
Contributors and the agent that ran the item are on the card, so results are attributable without asking in chat.
One click narrows the board to a colleague, or back to everyone. The filter follows you across devices.
Members who can write get a card only they can see, so personal work sits on the same list without being published to the team.
Not one assistant. As many as you have kinds of work
An agent here is a configuration, not a personality: its own model, its own system prompt, the devices it may use, and a permission for every tool it can call. Make one per kind of work and choose which one takes an item. They run in parallel, each on its own todo. Agents belong to the person who configured them: personal machines and logged-in browsers stay private, while the team’s cloud desktop is shared by everyone holding a seat.
This is the whole panel from the app, read-only on example data: the name, the model, the system prompt, and a permission on every tool — allow, ask first, or block.
Model and thinking level per agent, from every vendor on one plan. A cheap fast one for triage, a strong one for the code.
A system prompt per agent, so the finance agent and the marketing agent do not share one compromise prompt.
Which machines it can touch, which tools it may run, and which of those need a human to approve first.
A project per part of the company
Companies do more than one kind of work, so they run more than one project: engineering, marketing, support, finance, one per client. Each project has its own members, its own board and its own brand context, and remembers which agent you work it with, so switching projects from the picker switches the whole setup.
That is the difference from a tool that lives in one person's editor or terminal. Four departments end up on one system with one bill, instead of four tools that do not know about each other.
Roles for people, permissions for agents
Access to a project is read, write or owner. What an agent may do is separate and finer: every tool is allow, ask first, or block, per agent. Ask-first is a real gate — the call waits for a human, it is not a preference the agent can talk itself out of.
Secrets live in an encrypted vault rather than in prompts, the team cloud desktop is the only machine a seat shares, and any run can be stopped mid-task. Set everything that writes to production to ask first and widen from there.
Seats fund one pool, not one allowance each
Team seats cost $30 for a standard seat and $120 for a premium one, and the seat price goes into a single shared usage pool rather than a personal quota. 2 standard seats plus 1 premium seat is $180 a month and a $230 pool — the seats contribute $180 and we add $50.
So a colleague who has a heavy week draws more of the pool and a quiet week wastes nothing. The pool is genuinely shared: it is one balance for the team, and per-seat session and weekly windows pace how fast any one person can draw it down, with premium seats getting the larger windows. Usage is metered at 1.5× model cost.