Conversation

An AI chat transcript: role-attributed messages in a scrollable column that follows new replies without yanking a reader who scrolled up. For AI assistant panels.

Components

Conversation

feedback

interactive

<arc-conversation>

Overview

Conversation is the transcript surface for an AI assistant panel. Slot `<arc-message>` children into it and each one takes a voice from its speaker attribute: user messages align to the inline end on a faint accent tint, assistant messages answer from the inline start on a neutral surface, and system messages run centered and muted for notices in the transcript's own voice. Because alignment is built on logical properties, the whole layout mirrors in RTL. The container is the scroll area, and it follows one rule: while the reader is near the bottom, new or growing messages keep the view pinned to the latest; once they scroll up to re-read, the transcript stays put. The `arc-scroll-away` and `arc-scroll-return` events report those transitions with the distance from the bottom, so a consumer can float a "jump to latest" chip and wire it to the `scrollToEnd()` method. Streaming needs no API of its own: a message body is its default slot, so appending text to the slot streams the reply in. With the `markdown` attribute set, the slotted text renders through the markdown renderer and re-parses as it grows, and a `pending` message shows the typing indicator until the first tokens arrive. The transcript server-renders in full, typing dots included.

Guidelines

When to use

Use Conversation for a dialogue between the user and a responder, where role attribution and scroll behavior matter
Stream a reply by appending to the message's slotted text (for example, updating textContent as tokens arrive); with markdown set it re-renders as it grows
Show a pending assistant message the moment a request is sent, then fill its slot when tokens arrive
Give messages timestamps as ISO strings and let the relative-time rendering do the rest
Listen for arc-scroll-away to float a "jump to latest" chip, and clear it on arc-scroll-return
Size the transcript with the --conversation-height custom property, or let it fill a sized parent

When not to use

Do not use Conversation for an activity feed or event history. That is Timeline's job, where entries mark moments rather than speakers
Do not use it as a general item column. A List handles collections that no one is talking to
Do not scroll the reader to the bottom yourself on new messages; the component already follows, and only when the reader wants it to
Do not put critical status in a system message alone; it scrolls away with the transcript
Do not add avatars by reflex; the avatar slot is there for products that need one

Features

Three message voices: user on an accent tint at the inline end, assistant on a neutral surface, system centered and muted
Auto-scroll follows new and growing messages, but only while the reader is already near the bottom
`arc-scroll-away` / `arc-scroll-return` events with the distance from the bottom, for a "jump to latest" chip
`scrollToEnd()` method to jump the transcript to its newest message
Streaming-friendly: append text to a message slot and the view keeps up
Markdown rendering of message bodies through `arc-markdown`, re-parsed as streamed text grows
Typing indicator on `pending` messages, replaced by a static ellipsis under `prefers-reduced-motion`
Relative timestamps through `arc-time-ago`, with the absolute date on hover
An `avatar` slot on each message for products that want faces; none are built in
RTL mirrors automatically: alignment is logical-properties only
Server-renders in full, pending dots included

Preview

Project "Neon Tides" opened — the assistant can see your arrangement.

How do I loop the chorus while I try out lead takes?

Drag the **loop brace** over bars 17-24, then: 1. Press L to switch looping on 2. Arm the lead track 3. Hit record — every pass lands on its own take lane

Can it keep every take, or just the last one?

Usage

This component requires JavaScript. No pure HTML/CSS version is available. Use the Web Component directly or a framework wrapper.

<arc-conversation style="--conversation-height: 400px">
  <arc-message speaker="user" author="You" timestamp="2026-07-31T14:02:00Z">
    How do I loop the chorus while I try out lead takes?
  </arc-message>
  <arc-message speaker="assistant" author="daw[n]" markdown timestamp="2026-07-31T14:02:04Z">
    Drag the **loop brace** over bars 17-24, press L, then arm the
    lead track — every pass lands on its own take lane.
  </arc-message>
  <arc-message speaker="assistant" author="daw[n]" pending></arc-message>
</arc-conversation>

<script>
  // Streaming: fill the pending message as tokens arrive.
  const reply = document.querySelector('arc-message[pending]');
  function onToken(token) {
    reply.pending = false;
    reply.markdown = true;
    reply.textContent += token;
  }
</script>

import { Conversation, Message } from '@arclux/arc-ui-react';

export default function AssistPanel({ messages, streaming }) {
  return (
    <Conversation style={{ '--conversation-height': '400px' }}>
      {messages.map((m) => (
        <Message key={m.id} speaker={m.role} author={m.author} timestamp={m.at} markdown>
          {m.text}
        </Message>
      ))}
      {streaming && <Message speaker="assistant" author="daw[n]" pending />}
    </Conversation>
  );
}

<script setup>
import { Conversation, Message } from '@arclux/arc-ui-vue';
defineProps(['messages', 'streaming']);
</script>

<template>
  <Conversation style="--conversation-height: 400px">
    <Message
      v-for="m in messages"
      :key="m.id"
      :speaker="m.role"
      :author="m.author"
      :timestamp="m.at"
      markdown
    >{{ m.text }}</Message>
    <Message v-if="streaming" speaker="assistant" author="daw[n]" pending />
  </Conversation>
</template>

<script>
  import { Conversation, Message } from '@arclux/arc-ui-svelte';
  export let messages = [];
  export let streaming = false;
</script>

<Conversation style="--conversation-height: 400px">
  {#each messages as m (m.id)}
    <Message speaker={m.role} author={m.author} timestamp={m.at} markdown>{m.text}</Message>
  {/each}
  {#if streaming}
    <Message speaker="assistant" author="daw[n]" pending />
  {/if}
</Conversation>

import { Component, Input } from '@angular/core';
import { Conversation, Message } from '@arclux/arc-ui-angular';

@Component({
  imports: [Conversation, Message],
  template: `
    <arc-conversation style="--conversation-height: 400px">
      @for (m of messages; track m.id) {
        <arc-message [attr.speaker]="m.role" [attr.author]="m.author" [attr.timestamp]="m.at" markdown>
          {{ m.text }}
        </arc-message>
      }
      @if (streaming) {
        <arc-message speaker="assistant" author="daw[n]" pending></arc-message>
      }
    </arc-conversation>
  `,
})
export class AssistPanelComponent {
  @Input() messages: Array<{ id: string; role: string; author: string; at: string; text: string }> = [];
  @Input() streaming = false;
}

import { Conversation, Message } from '@arclux/arc-ui-solid';
import { For, Show } from 'solid-js';

export default function AssistPanel(props) {
  return (
    <Conversation style={{ '--conversation-height': '400px' }}>
      <For each={props.messages}>
        {(m) => (
          <Message speaker={m.role} author={m.author} timestamp={m.at} markdown>
            {m.text}
          </Message>
        )}
      </For>
      <Show when={props.streaming}>
        <Message speaker="assistant" author="daw[n]" pending />
      </Show>
    </Conversation>
  );
}

import { Conversation, Message } from '@arclux/arc-ui-preact';

export default function AssistPanel({ messages, streaming }) {
  return (
    <Conversation style={{ '--conversation-height': '400px' }}>
      {messages.map((m) => (
        <Message key={m.id} speaker={m.role} author={m.author} timestamp={m.at} markdown>
          {m.text}
        </Message>
      ))}
      {streaming && <Message speaker="assistant" author="daw[n]" pending />}
    </Conversation>
  );
}

API

autoScrollboolean=true: Follow new content: when a message is added or grows while the reader is near the bottom, scroll to keep the latest visible. A reader who has scrolled up is never pulled back down. Defaults to true; set the property to false to leave scrolling entirely to the consumer.

Methods

scrollToEnd(): Scroll the transcript to its newest message.

Events

arc-scroll-awaydetail:{value: number}: Fired once when the reader scrolls up out of the near-bottom zone of the transcript. detail.value is the distance from the bottom in pixels.
arc-scroll-returndetail:{value: number}: Fired once when the reader comes back within the near-bottom zone. detail.value is the distance from the bottom in pixels.

Message

<arc-message>

One message in the transcript. The speaker attribute picks the voice (user, assistant, or system), author and timestamp fill the muted meta line, markdown renders the slotted text through the markdown renderer, and pending shows the typing indicator until a reply arrives.

authorstring='': Display name shown in the muted meta line above the bubble. Omit it and the meta line only appears when a timestamp is set.
timestampstring='': When the message was sent, as an ISO 8601 string. Rendered as house relative time ("3 minutes ago") through arc-time-ago, with the absolute date on its title.
speaker'user' | 'assistant' | 'system'='user': Whose message this is. "user" aligns to the inline end on an accent-tinted surface, "assistant" to the inline start on a neutral surface, and "system" runs centered and muted for notices in the transcript's own voice. An unrecognized value renders as "user".
pendingboolean=false: Renders the typing indicator (three pulsing dots) in place of the body while a reply is being produced. Under prefers-reduced-motion the dots give way to a static ellipsis.
markdownboolean=false: Render the slotted text through the house markdown renderer. The slot's text content is the source; it re-parses whenever the slot changes, so streaming into the slot streams through the renderer. When false, slotted content renders as-is.

Conversation

Overview

Guidelines

When to use

When not to use

Features

Preview

Usage

API

Methods

Events

Message

See Also

Docs

Project

Arclight