All components

Polymorphic button with six variants and four sizes.

Single-line text input with prefix/suffix slots.

Multi-line input with optional autosize.

Surface with header, content, and footer slots.

Small status indicator.

Circular avatar with fallback initials and status dot.

Horizontal or vertical divider.

Accessible modal dialog built on Radix.

Side-anchored panel (top/right/bottom/left).

Mobile-first bottom sheet with drag-to-dismiss.

Floating element anchored to a trigger.

Contextual hint with delay control.

Transient notification (Sonner).

Custom select control built on Radix.

Searchable select using cmdk.

Boolean input with indeterminate state.

Toggle switch.

Grouped exclusive choice.

Multi-item menu with submenus and shortcuts.

Switch between equivalent views.

Collapsible disclosure rows.

Cmd+K-style command palette (cmdk).

react-hook-form + zod wrapper with FormField, FormItem, FormMessage.

Drag-and-drop uploader for documents, images, and audio fed to an AI pipeline.

Sortable, filterable, paginated table for AI extractions, runs, and evaluation results.

User / assistant / system bubble with streaming cursor for chat interfaces.

Auto-resizing composer with attachments for chat and agent interfaces.

Source snippet with title, locator, and link — for RAG citations.

Color-coded badge for any AI extraction or classification confidence score.

Number, delta, and sparkline at a glance — for AI dashboards and monitoring.

Entity summary card with a circular score and quick actions.

Positive / neutral / negative readout for sentiment-analysis output.

Timeline entry with actor, action, and diff — for AI audit trails.

Checkbox row with assignee, due date, and status — for agent task lists.

Auto-classified inbox row with preview and draft response.

Dropdown to pick an AI provider (Anthropic, OpenAI, Google, Mistral, Ollama).

Models grouped by provider with capability icons, context window, and pricing.

Masked API key input with show/hide toggle and live validation status.

Sampling temperature slider with deterministic ↔ creative endpoints.

Nucleus sampling slider for top-p configuration.

Numeric input that knows the model output cap and warns on overflow.

Auto-resizing system prompt editor with variable detection and token estimate.

Bar chart of tokens consumed vs. context window.

Live USD cost meter that subscribes to the AI event bus.

Operational / degraded / down pill with optional latency.

Composed configuration card combining provider, model, sampling, and prompt.

Scrollable conversation surface with auto-scroll-to-bottom and a "scroll to latest" floating button.

Plain-text incremental renderer with a blinking cursor while tokens arrive.

Markdown renderer tuned for streaming — GFM, themed code blocks, trailing cursor.

Code block with language badge, filename header, copy button, and live cursor.

Three-dot bouncer for "AI is typing" — subtle and bubble variants.

Hover-revealed copy / regenerate / edit / delete / fork / share row.

Live token estimate with optional context-window bar; tone shifts at 70% / 90%.

Left-rail conversation list with search, pinned items, and unread counts.

Tree visualisation of branched conversations — regenerations and edits.

Collapsible disclosure for the model’s extended-thinking output, with live streaming.

Vertical stepper for explicit, structured reasoning steps.

Lightweight "model is working" affordance — distinct from the typing indicator.

Card representing a single tool invocation by the model with status, args, and timing.

Auto-formatted output viewer — JSON / text / markdown / image — with copy and truncate.

Schema-driven form for tool parameters — string, number, enum, boolean, list, and JSON.

Searchable, groupable catalog of available tools with per-tool toggles.

Stacked timeline of tool invocations with status, args, result, and duration.

Card for a single configured MCP server — transport, status, capabilities, inline actions.

Multi-server view with search, "Add server" affordance, and connect/disconnect handlers.

Pill per MCP capability (tools, prompts, resources, sampling, roots, logging) with icons.

Inline pill describing the MCP connection lifecycle with optional latency readout.

Searchable list of resources exposed by MCP servers, optionally grouped by URI scheme.

Catalog of MCP prompt templates with collapsible argument lists.

Real-time JSON-RPC traffic viewer — direction badges, level pills, expandable JSON payloads.

Pill describing an agent lifecycle (idle/thinking/working/blocked/done/errored) with animated indicator.

Single-agent card with avatar, name, role, model, status, and the tools available to it.

Multi-agent team view with search, status filters, and list/grid layouts.

Vertical timeline of agent activity — thoughts, actions, tool calls, messages, handoffs, errors.

Card visualising a handoff between two agents with reason and pending/accepted/rejected state.

Hierarchical task tree with per-node status, assigned-agent chips, and progress bars.

Image bubble for chat replies with loading, generation progress, error, and ready states.

Responsive image grid with built-in lightbox, keyboard navigation, and download.

Bar-based audio waveform: live recording, playback progress, or idle baseline.

Compact player for TTS / transcribed audio — play, seek, mute, cycle speed, download.

Time-anchored transcript with active-segment highlight, auto-scroll, click-to-seek.

Vision-input dropzone — drag-and-drop, file picker, paste, optional camera capture.

Saved-prompt card with name, description, version, model, variable count, and tags.

Auto-form for `{{variables}}` extracted from a prompt template, with live preview.

Single eval metric with delta vs baseline and an optional inline sparkline.

Eval-run summary — status, prompt, model, dataset, progress, headline metrics.

Stacked table for an eval dataset — input / expected / actual / score with bucket filters.

Side-by-side comparison of two prompts (or two models) with metric deltas and sample outputs.

Single retrieved chunk with rank, source, score, optional reranker delta, and expandable metadata.

List of retrieved chunks with in-set search and a score-threshold slider.

Query input with topK slider, similarity threshold, and reranker toggle.

Source document with chunk boundaries highlighted — preview a chunking strategy.

Pure-SVG 2D scatter for UMAP / t-SNE / PCA projections with hover tooltips and group colour-coding.

Connected stage chips for embed → retrieve → rerank → generate with status, durations, and counts.

Single skill with icon, name, version, author, description, scopes, auth pill, and an enable toggle.

Pill row of required scopes (read / write / admin × resource) with compact and limited variants.

Auth lifecycle card — connected / expired / needs-reauth / never / errored — with primary action.

Installed-skills catalog with search, category filters, and per-skill enable toggles.

Vertical timeline of skill calls with status, action, duration, and expandable payloads.

Discovery grid with category filters, star ratings, install counts, and stateful install button.

Twinkling four-point stars at random positions inside a container — the "magic AI" affordance.

Animated conic-gradient border that orbits around an element — indicates "AI-touched" content.

Siri-style breathing orb with idle / thinking / speaking / errored states.

Animated network of nodes and edges drifting across a hero — pure canvas, GPU-accelerated.

Visual representation of tokens flowing through a horizontal lane — streaming indicator.

Soft, rotating conic-gradient glow behind any child element — element-scoped backdrop.