Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.
Go agentagenticaichatbotdsh-pluginembeddingsevaluationgenerative-aigolangknowledge-basellmmulti-tenantollamaopenaiquestion-answeringragrerankingsemantic-searchvector-searchwiki
<template>
<div class="embed-bot-msg" :class="{ 'is-embedded': embeddedMode }">
<div v-if="session?.isRagMode" class="rag-answer-stack">
<RagPipelineProgress :session="session" embedded-mode />
<AgentStreamDisplay v-if="session?.isAgentMode" :session="session" :session-id="sessionId" :user-query="userQuery"
rag-mode :embedded-mode="embeddedMode" :embed-channel-id="embedChannelId" :embed-token="embedToken"
:embed-session-sig="embedSessionSig" :embed-visitor-id="embedVisitorId" />
</div>
<template v-else>
<DocInfo v-if="session?.knowledge_references?.length" :session="session" embedded-mode />
<AgentStreamDisplay v-if="session?.isAgentMode" :session="session" :session-id="sessionId" :user-query="userQuery"
:embedded-mode="embeddedMode" :embed-channel-id="embedChannelId" :embed-token="embedToken"
:embed-session-sig="embedSessionSig" :embed-visitor-id="embedVisitorId" />
</template>
<DeepThink v-if="session?.showThink && !session?.isAgentMode" :deep-session="session" />
<div v-if="!session?.hideContent && !session?.isAgentMode" ref="parentMd">
<div v-if="hasActualContent" class="content-wrapper">
<div class="ai-markdown-template markdown-content" v-stable-html="renderedHTML" />
</div>
</div>
<Teleport to="body">
<div v-if="citationFloat.visible" class="embed-citation-float"
:style="{ top: `${citationFloat.top}px`, left: `${citationFloat.left}px` }" @mouseenter="cancelCitationClose"
@mouseleave="scheduleCitationClose">
<template v-if="citationFloat.type === 'web'">
<div class="embed-citation-float__title">{{ citationFloat.title || citationFloat.url }}</div>
<a v-if="citationFloat.url" class="embed-citation-float__link" :href="citationFloat.url" target="_blank"
rel="noopener noreferrer">{{ citationFloat.url }}</a>
</template>
<template v-else>
<div class="embed-citation-float__title">{{ citationFloat.title }}</div>
<div v-if="citationFloat.loading" class="embed-citation-float__muted">…</div>
<div v-else-if="citationFloat.error" class="embed-citation-float__error">{{ citationFloat.error }}</div>
<div v-else class="embed-citation-float__body">{{ citationFloat.content }}</div>
</template>
</div>
</Teleport>
</div>
</template>
<script setup lang="ts">
import { computed, defineAsyncComponent, nextTick, onMounted, onUpdated, ref, watch } from 'vue'
import 'katex/dist/katex.min.css'
import {
sanitizeMarkdownHTML,
safeMarkdownToHTML,
createSafeImage,
isValidImageURL,
hydrateProtectedFileImages,
} from '@/utils/security'
import type { ProtectedFileAccessContext } from '@/utils/protectedFileAccess'
import {
createChatMarkdownRenderer,
renderChatMarkdown,
} from '@/utils/chatMarkdownRenderer'
import {
createMermaidCodeRenderer,
ensureMermaidInitialized,
enhanceMarkdownContainer,
} from '@/utils/mermaidShared'
import { useEmbedCitationPopover } from '@/composables/useEmbedCitationPopover'
import { useTypewriter } from '@/composables/useTypewriter'
import { vStableHtml } from '@/directives/stableHtml'
const RagPipelineProgress = defineAsyncComponent(
() => import('@/views/chat/components/RagPipelineProgress.vue'),
)
const AgentStreamDisplay = defineAsyncComponent(
() => import('@/views/chat/components/AgentStreamDisplay.vue'),
)
const DocInfo = defineAsyncComponent(
() => import('@/views/chat/components/docInfo.vue'),
)
const DeepThink = defineAsyncComponent(
() => import('@/views/chat/components/deepThink.vue'),
)
ensureMermaidInitialized()
const markdownRenderer = createChatMarkdownRenderer({
codeRenderer: createMermaidCodeRenderer('mermaid-embed-botmsg'),
imageRenderer: ({ href, title, text }) => createSafeImage(href, text || '', title || ''),
isValidImageUrl: isValidImageURL,
})
type EmbedSession = {
content?: string
isRagMode?: boolean
isAgentMode?: boolean
showThink?: boolean
hideContent?: boolean
is_completed?: boolean
agentEventStream?: Array<Record<string, unknown>>
knowledge_references?: Array<{ chunk_type?: string; knowledge_id?: string; knowledge_title?: string }>
}
const props = withDefaults(
defineProps<{
content?: string
session?: EmbedSession
sessionId?: string
userQuery?: string
embeddedMode?: boolean
embedChannelId?: string
embedToken?: string
embedSessionSig?: string
embedVisitorId?: string
}>(),
{
content: '',
session: () => ({}),
sessionId: '',
userQuery: '',
embeddedMode: true,
embedChannelId: '',
embedToken: '',
embedSessionSig: '',
embedVisitorId: '',
},
)
const parentMd = ref<HTMLElement | null>(null)
const embedChannelIdRef = computed(() => props.embedChannelId)
const embedTokenRef = computed(() => props.embedToken)
const {
float: citationFloat,
rebind: rebindCitations,
cancelClose: cancelCitationClose,
scheduleClose: scheduleCitationClose,
} = useEmbedCitationPopover(
parentMd,
embedChannelIdRef,
embedTokenRef,
{
getKnowledgeReferences: () => props.session?.knowledge_references,
},
)
// Smooth the streamed answer into a steady typewriter cadence (shared with the
// Agent path). History reloads arrive complete and snap to full.
const answerText = computed(() => String(props.content || props.session?.content || ''))
const { displayed: typedAnswer } = useTypewriter(
() => answerText.value,
() => Boolean(props.session?.is_completed),
)
const renderedHTML = computed(() => {
const text = typedAnswer.value
if (!text.trim()) return ''
return renderChatMarkdown(text, {
renderer: markdownRenderer,
escapeMarkdown: safeMarkdownToHTML,
sanitizeHtml: sanitizeMarkdownHTML,
streaming: !props.session?.is_completed,
})
})
const hasActualContent = computed(() => {
const text = String(props.content || props.session?.content || '')
return text.trim().length > 0
})
const hydrateImages = async () => {
const embedAccess: ProtectedFileAccessContext | undefined =
props.embedChannelId && props.embedToken
? { mode: 'embed', channelId: props.embedChannelId, token: props.embedToken }
: undefined
await hydrateProtectedFileImages(parentMd.value, embedAccess)
}
const renderMermaidDiagrams = async () => {
await enhanceMarkdownContainer(parentMd.value)
}
watch(renderedHTML, () => {
nextTick(async () => {
rebindCitations()
await hydrateImages()
if (props.session?.is_completed) {
await renderMermaidDiagrams()
}
})
})
onUpdated(() => {
nextTick(async () => {
await hydrateImages()
if (props.session?.is_completed) {
await renderMermaidDiagrams()
}
})
})
onMounted(() => {
nextTick(async () => {
await hydrateImages()
await renderMermaidDiagrams()
})
})
</script>
<style scoped lang="less">
@import '../../components/css/chat-markdown.less';
@import '../../components/css/chat-citations.less';
.embed-bot-msg {
border-radius: var(--app-radius-xs);
color: var(--td-text-color-primary);
font-size: var(--app-text-xl);
margin-right: auto;
max-width: 100%;
box-sizing: border-box;
&.is-embedded {
width: 100%;
:deep(.agent-stream-display) {
width: 100%;
}
}
}
.rag-answer-stack {
display: flex;
flex-direction: column;
gap: 0;
}
.content-wrapper {
padding: 2px 0;
}
.markdown-content {
// Chat Markdown visual styles are centralized in chat-markdown.less.
// Do not add element-level Markdown rules here; update the shared mixin.
.chat-markdown-typography();
.chat-citation-pills();
}
.embed-citation-float {
position: absolute;
z-index: 10000;
max-width: 320px;
padding: 10px 12px;
border-radius: var(--app-radius-md);
background: var(--td-bg-color-container);
box-shadow: 0 6px 18px rgba(0, 0, 0, 0.18);
font-size: var(--app-text-sm);
line-height: 1.5;
color: var(--td-text-color-primary);
&__title {
font-weight: 600;
color: var(--td-brand-color);
margin-bottom: 4px;
}
&__link {
color: var(--td-brand-color);
word-break: break-all;
}
&__body {
max-height: 200px;
overflow-y: auto;
white-space: pre-wrap;
}
&__muted {
color: var(--td-text-color-placeholder);
}
&__error {
color: var(--td-error-color);
}
}
</style>