React components for supervising AI agents: approvals, tool calls, run timelines, diff review and handoffs. 109 components, light and dark.
The point where the agent hands control back: outcome, summary, next actions.
The agent event model, useAgentRun and its selectors: turn events from any agent into run state for the agent components.
The run screen. It takes organisms as slots, so your app decides what goes where.
A message in the page: what happened and what to do.
One specific action to allow or deny. A, Shift+A and D work while focus is inside it.
The frame of an app: sidebar, header, content and an optional aside.
Agents are rounded squares; people are circles.
A status as tone, icon and word.
A full-width notice in three tones.
Amounts across categories. More than one series stacks.
A ranking: a name, a bar and a value on every row.
Where you are. Long trails fold the middle.
A browser window drawn around a preview.
Spend against a limit. It warns at 80% and flags going over.
A glass capsule. Blue is for the one primary action; red text marks refusal.
A day or a range. Arrow keys move by day, Page Up and Down by month.
A glass tile for grouping related content.
A whole conversation with an agent: messages, suggestions and the input. Type something to try it.
A native checkbox with its label.
Pick a point to roll back to and see what it undoes before confirming.
Card-sized options for a choice that needs a sentence each.
A text field with a clear button that puts focus back in the field.
Inline code, or a scrollable block.
Sections that open and close.
Type to filter and pick one. Arrow keys move, Enter picks, Escape closes.
Search everything and act from the keyboard. Press ⌘K or Ctrl+K to see it in use: the search on this site is a CommandMenu.
How sure the agent is: a range in words with reasons, never a percentage.
A preview on hover or keyboard focus. Focus never moves into it.
Opens on right-click, long-press, or Shift+F10.
Copies text, shows a check for two seconds, and tells screen readers.
Shared helpers, base styles and the design tokens every Creance component uses.
Shared layout and styles for the chart components.
Terms and their values.
For actions that cannot be undone. The red button stays off until the name is typed.
A question over a dimmed page. It traps focus and returns it when closed.
One line of a diff, with line numbers and screen-reader text.
A decision per hunk before anything lands. J and K move, Y and N decide.
Parts of a whole, with each slice’s value and share in the legend.
A "…" button with secondary actions for one row.
An empty list as an invitation to act.
An agent or a person as a row.
A failure that needs a decision: what broke, the error, and what you can do.
What went wrong and what to do next.
Thumbs up or down, then an optional comment.
A label, a control, and the hint or error that describes it, wired for screen readers.
Files and folders with change markers. Arrow keys walk the tree.
Many agents at once, as birds on lines back to you. The one that needs you flies in on a taut line.
The end of a page: brand, groups of links, and a line for copyright.