Insights
The home a workspace opens on at 9:00: what its agents did overnight, what needs you, and what comes next. One night told three ways: as a brief, as a queue of decisions and as an operations report.
One night, three readings#
The three screens report the same fifteen hours from the same six agents, and their numbers agree: 412 tasks, 94% of them handled without a person, five things waiting on you. The brief reads like a letter that ends, for whoever wants the story first. The decisions screen is a queue to empty, for whoever approves. The operations screen counts every agent by outcome against the night before, for whoever runs the agents.
What makes a report worth trusting#
- What needs you comes before what got done. A blocked agent is never buried under its own wins.
- Every claim in the written summary carries a
Citationto the ticket, refund or deploy it comes from. - The report says when it was written and what it compares against, once, at the top.
- Agents are counted by outcome, handled or handed off, not by how busy they were.
- Doubt stays visible: a finding the agent could not settle is marked inconclusive, not guessed.
- Nothing a recommendation proposes runs until you approve it, and every decision can be undone.
Built from#
Every screen is the docked Layout with the workspace Sidebar, the header bar and a PageHeader whose heading states the night's verdict. Numbers are Stat parts, section headings are PageSectionTitles with their explanation in a tooltip, and the written summary anchors its claims with CitationClaim. The brief folds each agent's night into a minimal Accordion and closes on a PromptInput with starters for a follow-up question.
Responsive#
Each preview renders inside an iframe, so the breakpoints respond to the frame width. Below lg the rail moves into the drawer behind the mobile bar, number bands fold to two columns, side columns drop under the main one, and tables scroll sideways inside their own frame.