The AI Overview root, built on shadcn Collapsible and Button, with a collapsible Show more.
A source chip built on shadcn Badge and HoverCard.
Showing 49–72 of 357
The AI Overview root, built on shadcn Collapsible and Button, with a collapsible Show more.
A source chip built on shadcn Badge and HoverCard.
Showing 49–72 of 357
Paragraph, heading, list and link styles for AI Overview answers.
A mark that sweeps in Google's blue highlight.
AI chat input with a model menu, a separate Effort control, file chips you can drop or paste, a / command menu, and Stop while replying.
Voice recorder pill that morphs into a live waveform with timer, pause, cancel and stop, then a scrubbable playback row.
Masks names, emails, card numbers and keys as placeholders before a prompt is sent, then fills the real values back into the reply.
Renders a streaming assistant response supplied by your app — text, fenced code with copy, inline [n] citations to a sources rail, and streaming/stopped/complete/error states with Stop, Retry, and Copy. Presentation only: it never calls a model. Reduced-motion shows text instantly; a polite live region announces completion, not tokens.
Meeting times the assistant found, each with a strip of both calendars so you can see why it is free. Booking folds the list into the event.
An email the assistant drafted, editable in place. Send starts a short countdown you can undo before anything goes out.
AI Summary: the banner that marks AI-authored copy — one sentence of plain language on an animated vector field, in one of four palettes. The painting is a gradient plus colour masses under a fractal-noise warp, so a palette is a few hex values and three durations. No runtime dependencies.
Shows an AI working before its answer — a shimmering "Thinking" with what it is doing — then folds into one quiet "Thought for 8 s" row that opens the reasoning.
AI loading states: shimmer text, bouncing dots, glowing orb, collapsible reasoning steps and simulated token streaming.
An answer where every unsure word shows what else the model weighed. Pick another word and the reply writes itself again from there.
Approve or deny an agent's tool calls. Destructive calls need a press and hold, and decided calls fold into a trace with their output.
Tool calls inside a chat reply. Each call streams in as one row with a live timer, then they fold into a single line once the answer is ready.
Tabbed AI prompt input with assistant, blog and pro modes plus file upload.
Voice input with a scrolling waveform, then a take you can play back, discard or send.
Voice activity visualizer with orb, bars, wave and dots variants, driven by the microphone, simulated speech or a controlled level.
AI writing assistant panel with assistant, blog and pro modes and a prompt field.
AlimTalk (알림톡) preview mirroring the API fields — messageType BA/EX/AD/MI, emphasizeType NONE/TEXT/IMAGE/ITEM_LIST, buttons, quick replies, length limits.
An action an agent wants to take, waiting for a person. Say what it does and what can't be undone; press and hold the brass seal to approve.
A composable decision surface that keeps risk, evidence, and impact explicit.
The approval card as a lifecycle: a decision that survives a closed tab, is idempotent across two open ones, and says what happened when it lands after the request expired.