5.0Monitor
See what moved, stalled or failed
Open a daily brief, then trace any result through its inputs, tool calls, handoffs and decisions.
W30W41
Atlas3h 10m
Mira1h 46m
You3h 52m
Illustrative figures. Your workspace shows its live run data.
This week, so far.
Two lines run out this week
Ceramic planter and wool throw. Reorders drafted, waiting on you.
Courier claims settled
Four closed since Friday. One is still open with the carrier.
Ardleigh has not confirmed
Third chase sent. Vela has a call prepared if you want it made.
Click any bar to read that week's brief. The tallest one is the week it went wrong.
W41. This week, so far.Transcripts are workspace data. They are visible to the people in your workspace and are not used to train anyone's model — see the security page for what is retained and for how long.
The brief
Monday morning, in about a minute
Analytics that require you to go and look at them do not get looked at. The brief comes to you: a short account of what moved, what is stuck and what is waiting on you, in the order those things actually matter, with a link into each one.
- 5.1Ordered by consequenceItems are ordered by consequence. The one most likely to cost money if ignored appears first with the reason stated.
- 5.2What is on youApprovals waiting on a person are separated from work the agents are still doing, so you can see your own list without reading everyone else's.
- 5.3Where you want itOn the dashboard, and by mail on the schedule you choose. Weekly if daily is too much.
Analytics
Which part of the week is the problem
Volume, time to settle and queue time are split by agent and by person. The figures show which stage of each job consumes the most time.
- 5.4Time to settleSee how long items take from arrival to close, by board and type, including the full distribution and its slowest cases.
- 5.5Where it queuesWhich column items sit in longest. A queue in front of a person is a different problem from a queue in front of an agent, and they are shown apart.
- 5.6By agent, by personThe same figures split both ways, so 'the agents are slow' and 'the approvals are slow' cannot be mistaken for each other.
- 5.7Spend and allowanceWhat has been approved, what runs and voice minutes have drawn from the plan, and how much of the allowance is left — before anything beyond it is charged.
Transcripts
Read the complete run
Every run keeps what the agent read, which tools it called, what came back and what it concluded. Open that record when you need to understand a result.
- 5.8The full runInputs, tool calls, responses and the decision. Nothing elided, and readable by anyone in the workspace who can see the item.
- 5.9Fix it at the sourceThe transcript links back to the agent's prompt and scope so you can correct the source of a bad run.
- 5.10Team chatA persisted conversation space for the people in the workspace, kept deliberately separate from the agents' own working record.
The rest of the system
- 1.0IntakeMail, marketplace messages and stock gaps become items on a board, already routed.
- 2.0WatchCover counted against what is selling, so a line that runs out is flagged early.
- 3.0ActNamed agents work the item and hand it between themselves with the evidence attached.
- 4.0ApproveAnything irreversible or financial is written down and held for a named person.
- 8.0MarketplaceAgents other operators built, installed with their tools declared before they run.
- 9.0LimitsA ceiling on what runs unattended, and the actions no ceiling ever covers.
- 10.0BoardsOne board held by people and agents alike, where the column follows the owner.
- 11.0PeopleRoles for the humans, a deliberately weaker one for the agents, one permission model.
- 12.0WorkflowsSchedules and events that start work on their own, each pausable without stopping the rest.
Start with one agent and one job.