Accordion that shows each answer as a chat bubble from a support avatar.
Accordion that shows each answer as a chat bubble from a support avatar.
Chat-style accordion that shows typing dots before the answer bubble appears.
Chat-style accordion where each question is a right-aligned user bubble and the answer replies on the left.
Chat-style answers with thumbs up and thumbs down buttons that toggle.
Approval gate for a proposed assistant bulk change showing sample before and after field values with approve and reject buttons.
Live huddle surface with elapsed timer, presenting banner, participant tiles showing mute and speaking states, and a call control bar.
Live agent run with tool calls, progress, and a stop action.
An agent turn with four tool calls waiting for approval, from a low-risk message that approves with undo to a critical delete that asks for the project name. Each call shows its tool, risk and arguments, then its result once it runs.
A chat with an agent as one screen, from the first word to the deploy: the conversation, the thinking, a plan, a run of tool calls, an approval, the answer with its code and the composer. A scripted run plays every state without a model and waits for you at the approval.
A run as three lanes: when an agent hands in its work the card travels to Needs review, Approve sends it on to Done, Request changes sends it back with a note, and cards also move by drag or from their menu.
One agent on a card, with its model, instructions, tools and knowledge files that upload, and a Test tab that answers: Publish opens what changed and three sample prompts that run, fails once, then Retry draws the check of the new live version.
Someone says what should happen, the assistant checks off its work in the chat and three cards type their own fields on a canvas, one after the other, until Run it becomes Live.
AI assistant thread that types a request, thinks over a sparkle field, asks a follow-up with a select and choice tiles, then streams its answer word by word.
A conversation that thinks, reads its sources and streams, with the plan checking itself off beside it: when the agent needs a yes, the field you were typing in becomes the request and turns back once you answer.
Example usage of AI.
Example implementation of agent.
Example usage of Agent Fleet.
Example usage of Agent Handoff.
Example usage of Agent Inbox.
Composed Pearl assistant turn: thinking, timeline, todos, tools, task card, markdown, citations, follow-ups.
A Sundermere operations page on one run: a header that condenses into a slim bar as the page scrolls, a live graph where an edge draws itself and a dot travels out and back for each call, a monitor with the command that waits for Allow or Deny, a budget bar that passes its 80% mark, a fleet that makes every panel follow the agent you pick, a trace waterfall and a board whose cards travel between lanes, in three columns that fold to one by container width. Plug your backend into onApprove and onDeny.
An editable hero block for AI/automation products that shows a real agent workflow with outcome copy + CTAs. Composes Agent Run Timeline (trimmed), Prompt Composer (compact), Tool Call Activity (one active tool), and Source Citation Rail (one result) at reduced complexity across idle/running/tool-active/waiting-for-approval/completed/failed. App-controlled — never simulates a model or shows chain-of-thought. SSR-safe, reduced-motion-safe. Clearly-fictional demo data.
An editable plan checklist the agent proposes before running.
A floating chat-bubble trigger that expands into a compact agent session popup, suitable for embedding on any page.
Showing 1–24 of 468