Two screens under a workspace's Settings, and one for your agency. They have one thing in common: you decide how much the machine does by itself.
Notifications
The platform reports what happens itself: a workspace was created, the status changed, an invitation was accepted, the brand was confirmed, a brand guide or logo package is ready, the monthly cap is at 80 or 100 percent, a demo ends, a connection no longer works. Per kind you choose what happens: a notification in the bell, a mail, both or nothing.
- An agency chooses for all its clients at once. A client workspace can deviate per kind; choose "Follow the agency or the default" to go along again.
- What the bell already shows live from the state (cap, demo, status, connection) is always in the bell. For those you only choose whether you also want a mail.
- The mail goes to the owners of the workspace, or of the agency when the agency's choice applies. An event gives exactly one mail per recipient.
Automatic
The autopilot has three levels.
| Level | What goes by itself |
|---|---|
| Suggest only (default) | Nothing. Every action waits for your signature. |
| Low risk by itself | Low-risk actions (R1) that cost nothing, for the channels you gave permission for. |
| Medium risk too, within budget | Also actions that affect behaviour (R2), as long as they stay within the monthly budget. |
High risk (R3) always goes to a person, at every level. The level of an action is fixed in the machine itself; a suggestion cannot choose its own level.
A level alone is not enough. Every channel (for example Meta or Google) has its own permission, and it is off by default. Permission for one channel says nothing about another. What may not go by itself appears as a suggestion under Approvals, with the same signature as everything else.
Only an owner or administrator changes the autopilot. Every change is in the audit trail, with the old and the new level.
The trail and reversing
Below the settings is what the machine has done: which action, at which level, by itself or with a signature, what it cost and how it ended. An executed action that the machine can reverse has a Reverse button. An owner or administrator does that, once, and it is recorded in the trail.
If the emergency stop is on, or the workspace is paused, the machine carries out nothing. What was already signed waits until it is allowed again.
Workflows: what the crew does by itself
Under Automation, Workflows shows every workflow of your workspace as a row of steps: the event that starts it, the context it only reads, the agent's thinking, the gate where a person signs and the action after it. You can only read here; there is no button that runs anything.
- Click a step to see what it does: its settings, the level (R1 is small and reversible) and whether a person signs before it. An action with no gate before it says so.
- Click a run in the list below to see the steps as they ran: done, working, waiting to be signed, failed, not done or locked. Each step shows its outcome, cost and time. A failed or refused run says which steps were therefore not run.
- Text a model wrote sits in a dashed box and is never run. A key or password never appears with its value, only its name.
- You see the last ten runs per workflow. Older ones are in the History.