For agencies and consultancies
A separate world for every client
The thing that breaks when agencies adopt AI is not quality. It is separation: whose context, whose credentials, whose data, and who approved what.
What is actually hard about this
Shared tooling leaks context
One workspace holding six clients means six brand voices in one context and, eventually, one client's positioning showing up in another's deck. It is not a hypothetical risk, it is a matter of time.
Credentials are the real blocker
Every client has their own CRM, their own social accounts and their own rules about who holds the keys. A tool with one integration slot per provider cannot serve an agency at all.
Approval is a commercial document
For an agency, who signed off which version is not admin. It is what you point at when a client says the work was not what they asked for.
What changes
The agents named below are the ones that actually do this work, from a fleet of 50.
One world per client, isolated
Separate context, separate knowledge, separate brand kit. Nothing crosses between them, so a client's positioning cannot surface in another client's work.
Multi-account integrations
Each client's HubSpot, Attio, Slack and Apollo connect under their own world with their own credentials, encrypted per tenant. No shared keys, no cross-client reach.
A review lane you can safely give the client
The client role sees the Dock and only the Dock: the work sent for review, their pins, their sign-off history. Not your chats, not your other clients, not the machinery.
Sign-off that stands up later
Every decision recorded against a specific version, with the notes that produced it, and unresolved notes carried into the next round. Scope arguments become a link rather than an email search.
A sensible first month
Not a rollout plan. Just the order that tends to work.
- 1Week oneBuild worlds for your two most different clients. Two different ones will teach you more than five similar ones.
- 2Week twoRun one real deliverable per client, all the way to client sign-off in the Dock. The point is testing the review lane with a real client, not testing the writing.
- 3Week threeConnect one client's CRM. Confirm for yourself that the credentials are scoped to that world before you connect a second.
- 4Week fourDecide what you will and will not put through this, and write it down. An agency that has not drawn that line will eventually be asked about it by a client.
Where most agencies start
The Agency plan
Twenty-five worlds and ten seats, which is a client roster rather than a pilot, plus the credit ceiling to run real production across all of them.
80,000 credits a month · 10 seats · 25 worlds. Your first world is free and needs no card.
Questions this raises
Can a client see anything except their own work?
No. The client role is restricted to the Dock for the world they were invited to. They cannot open chats, dispatch missions, see other worlds or reach the workspace settings.
Do we have to tell clients we use AI?
That is your call and your contract, and we take no position on it. What we can tell you is that the sign-off trail is designed to be shown to a client rather than hidden from one.
What happens to a client's world if we part ways?
Everything is exportable, and worlds are deletable individually. Client data is never used to train models, and deleting a world removes its context from every agent immediately.
Is there a white-label option?
Generated documents carry the client's brand, captured from their own site, rather than ours. The application itself is not white-labelled today, which is why the client role is scoped as tightly as it is.