Conversation & AI

Owner rule

Model output is never trusted as markup.

Size budget

39.3 kB gzip for the whole domain; it only goes down.

Entries

7 entries, 5 used by live products and sites. Experimental entries come last: they may change or go in any minor version.

• Analysis reportBetaThe result of an AI analysis, honest and calm: one sentence and an overall status, findings grouped by status with what was observed and what it may indicate, what the report did not examine, the next steps and where the result comes from.
• ChatBetaA conversation's parts — MessageBubble (time, receipt, retry), TypingIndicator, MessageComposer, ConversationItem — the same API on the web and React Native.5 files import it
• Engine flowBetaAn engine at the hub and the apps it serves around it, with what travels between them: photons going in (requests) and coming back (results) along each conduit — radial on a wide screen, top to bottom on a phone, planned apps dim and still.5 files import it
• Generated viewBetaDraws the interface a model describes in A2UI v0.9 — tables, figures, charts, cards, callouts — from a catalog the app declares, never from code.1 files import it
• MarkdownBetaA Markdown text on the tokens — headings, lists and tasks, quotes, tables, links, images, footnotes, fenced code in CodeBlock — parsed by TanStack Markdown, safe by default (raw HTML is never rendered), with a streaming mode for a text that arrives in pieces: an AI answer. React Native draws the same AST in Text and View.6 files import it
• Next actionsBetaTwo or three ranked, host-provided actions with their reason and urgency.1 files import it
• Proposal reviewExperimentalAn agent proposes, a person approves, rejects or edits: who proposes, what, how sure it is (in words and figures), the evidence as a signed diff, and the decision announced.