Collections
Items grouped by the job they do together. Open one to see them run and install them all at once.
- Human approvalThe agent asks before it sends, pays or deletes, the run waits for the answer, and a queue holds the asks when there are many.5 items: Approval card, Run timeline, Agent inbox, Dialog, Button
- Agent runsWhat a run is doing while it works, each step and call it made once it is done, and how its scores compare with the last one.5 items: Run timeline, Thinking state, Tool call row, Eval table, Drawer
- Streaming answersA model's answer arriving word by word at reading speed, with the thinking and the tool calls that came before it.4 items: Streaming text, Thinking state, Tool call row, Button
- Tool callsEvery call the model makes in plain words, with its timer, its input and output one click away, and an approval when the call is risky.4 items: Tool call row, Run timeline, Approval card, Drawer
- Overlays and promptsThe panels around the work: a decision in a dialog, details in a drawer, a quick change in a popover, a label on every icon button.5 items: Dialog, Drawer, Popover, Tooltip, Input