Registry

A composable AI chat input with stacked island panels, file attachments, suggestion chips, and context usage visualization.

Universal tool call component with running/completed/failed status states. Works standalone for sequential calls or inside ToolTree for parallel execution.

A composable tree visualization for parallel tool calls with L-connectors, expandable children, and hover timestamps. Wraps ToolCall children.

A composable item row for sources, memories, findings, files, tasks, and agent activity with media, metadata, and action slots.

The product-wide status vocabulary: one shape per state — check for complete, spinner for active, dashed circle for pending, clock for blocked, alert for warning and error — with sr-only labels.

A step disclosure pattern for showing visible agent work, touched sources, state, and completion without exposing hidden reasoning traces.

A composable confirmation surface for approvals, risky actions, clarifying interruptions, and rollback decisions.

A file attachment lifecycle pattern for drag states, validation, upload progress, rejection, retry, and removal.

A rich operational table for multiple agents with status, active task, progress, confidence, cost, and last activity.

A controlled question group for collecting missing requirements through text, single-choice, and multiple-choice prompts.

A locked consequence preview for approvals, external actions, document changes, and rollback-aware agent work.

A citation and provenance preview card with source title, excerpt, location, navigation, and open-source action.

A source-backed output document surface with cited sections, missing-source states, version metadata, and review status.

A compact token, cost, limit, and coverage meter for visible agent run budgets.

A responsive agent workbench that keeps chat visible while opening an inspectable browser, source, diff, or artifact surface beside it.

A compact settings summary showing the agent policy currently produced by autonomy, approval, notification, and memory controls.

A provenance-rich memory row with scope, expiry, last-used metadata, source inspection, edit, and removal actions.

A dense operational trace for background runs with source touches, status, warnings, timing metadata, and expandable recovery detail.

An ownership-transfer surface for multi-agent workflows with sender, receiver, payload, source basis, receiver action, and recovery path.

An end-to-end agentic review workflow block with observable work, source-scope clarification, locked action preview, and approval.

A human-in-the-loop approval block with observable preparation, locked action preview, policy reason, expiry, and decision surface.

A missing-decision collection block with structured answer options and a paused observable-work trace.

A provenance workflow block that pairs a source-backed output document with source previews and usage budget.

A memory review queue block with provenance, scope, expiry, source preview, and save/reject decision.

A background run monitoring block with an operational trace, visible run budget, agent status table, warnings, and blocked recovery path.

A multi-agent ownership transfer block with handoff packet, sender/receiver state, source basis, recovery path, trace, and agent status table.

A durable agent policy settings block with clickable switch labels and effective policy preview.