]*>[\s\S]*?<\/div>[\s\S]*?<\/div>/g;
+
+/** Inject trusted Mermaid SVG after DOMPurify, replacing the streaming loading shell. */
+export const injectCachedMermaidSvg = (
+ html: string,
+ cachedSvgHtml: string | null | undefined,
+): string => {
+ if (!html || !cachedSvgHtml) return html;
+ return html.replace(
+ STREAMING_MERMAID_LOADING_RE,
+ buildMermaidBlockHtml(cachedSvgHtml, 'data-mermaid="cached"'),
+ );
+};
+
export const formatManualTitle = (question?: string): string => {
if (!question) {
return i18n.global.t('chat.sessionExcerpt');
diff --git a/frontend/src/utils/citationChunkCache.test.mjs b/frontend/src/utils/citationChunkCache.test.mjs
new file mode 100644
index 00000000..b8ed94ae
--- /dev/null
+++ b/frontend/src/utils/citationChunkCache.test.mjs
@@ -0,0 +1,38 @@
+import assert from 'node:assert/strict'
+import test from 'node:test'
+import {
+ clearCitationChunkCache,
+ getCitationChunkCache,
+ makeCitationCacheKey,
+ setCitationChunkCache,
+} from './citationChunkCache.ts'
+
+test('makeCitationCacheKey combines scope and chunkId', () => {
+ assert.equal(makeCitationCacheKey('session-1', 'chunk-a'), 'session-1::chunk-a')
+ assert.equal(
+ makeCitationCacheKey('embed:ch:tok', 'chunk-b'),
+ 'embed:ch:tok::chunk-b',
+ )
+})
+
+test('get/set are isolated per scope', () => {
+ clearCitationChunkCache()
+ setCitationChunkCache('session-1', 'chunk-a', { content: 'one' })
+ setCitationChunkCache('session-2', 'chunk-a', { content: 'two' })
+
+ assert.equal(getCitationChunkCache('session-1', 'chunk-a')?.content, 'one')
+ assert.equal(getCitationChunkCache('session-2', 'chunk-a')?.content, 'two')
+})
+
+test('clearCitationChunkCache clears one scope or all', () => {
+ clearCitationChunkCache()
+ setCitationChunkCache('session-1', 'chunk-a', { content: 'one' })
+ setCitationChunkCache('session-2', 'chunk-a', { content: 'two' })
+
+ clearCitationChunkCache('session-1')
+ assert.equal(getCitationChunkCache('session-1', 'chunk-a'), undefined)
+ assert.equal(getCitationChunkCache('session-2', 'chunk-a')?.content, 'two')
+
+ clearCitationChunkCache()
+ assert.equal(getCitationChunkCache('session-2', 'chunk-a'), undefined)
+})
diff --git a/frontend/src/utils/citationChunkCache.ts b/frontend/src/utils/citationChunkCache.ts
new file mode 100644
index 00000000..1a847d60
--- /dev/null
+++ b/frontend/src/utils/citationChunkCache.ts
@@ -0,0 +1,35 @@
+export type CitationChunkCacheValue = { content: string; error?: string }
+
+const chunkCache = new Map
()
+
+export function makeCitationCacheKey(scope: string, chunkId: string): string {
+ return `${scope}::${chunkId}`
+}
+
+export function getCitationChunkCache(
+ scope: string,
+ chunkId: string,
+): CitationChunkCacheValue | undefined {
+ return chunkCache.get(makeCitationCacheKey(scope, chunkId))
+}
+
+export function setCitationChunkCache(
+ scope: string,
+ chunkId: string,
+ value: CitationChunkCacheValue,
+): void {
+ chunkCache.set(makeCitationCacheKey(scope, chunkId), value)
+}
+
+export function clearCitationChunkCache(scope?: string): void {
+ if (scope === undefined) {
+ chunkCache.clear()
+ return
+ }
+ const prefix = `${scope}::`
+ for (const key of chunkCache.keys()) {
+ if (key.startsWith(prefix)) {
+ chunkCache.delete(key)
+ }
+ }
+}
diff --git a/frontend/src/utils/citationMarkdown.ts b/frontend/src/utils/citationMarkdown.ts
index 8678a985..acf23e06 100644
--- a/frontend/src/utils/citationMarkdown.ts
+++ b/frontend/src/utils/citationMarkdown.ts
@@ -1,6 +1,22 @@
/** Shared citation tag preprocessing for chat markdown (QA + agent). */
+/** Self-closing or unclosed `` / `` tags from model output. */
+export const KB_WEB_TAG_RE = /<(?:kb|web)\b[^>]*?\s*\/?>/g
+const KB_TAG_ATTR_RE = /]*?)\s*\/?>/g
+const WEB_TAG_ATTR_RE = /]*?)\s*\/?>/g
+
const ATTRIBUTE_REGEX = /([\w-]+)\s*=\s*"([^"]*)"/g
+const UUID_RE = /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i
+
+export type CitationKnowledgeRef = {
+ id?: string
+ knowledge_id?: string
+ knowledge_title?: string
+ knowledge_filename?: string
+ chunk_index?: number
+ chunk_type?: string
+ knowledge_base_id?: string
+}
function parseTagAttributes(attrString: string): Record {
const attributes: Record = {}
@@ -30,12 +46,88 @@ function truncateMiddle(text: string, maxLength = 13): string {
return `${start}...${end}`
}
+function normalizeDocTitle(title: string): string {
+ return title.trim().toLowerCase()
+}
+
+function docTitlesMatch(a: string, b: string): boolean {
+ if (!a || !b) return false
+ const na = normalizeDocTitle(a)
+ const nb = normalizeDocTitle(b)
+ return na === nb || na.includes(nb) || nb.includes(na)
+}
+
+/** Map model context index (1, FAQ-1, DOC-2) to the real chunk UUID from retrieval results. */
+export function resolveCitationChunkId(
+ rawChunkId: string,
+ attrs: { doc?: string; kbId?: string },
+ refs?: CitationKnowledgeRef[] | null,
+): string {
+ const raw = String(rawChunkId || '').trim()
+ if (!raw || UUID_RE.test(raw)) return raw
+
+ const list = (refs || []).filter((r) => r && r.chunk_type !== 'web_search')
+ if (!list.length) return raw
+
+ const doc = (attrs.doc || '').trim()
+ const kbId = (attrs.kbId || '').trim()
+
+ if (doc) {
+ const byDoc = list.find(
+ (r) =>
+ docTitlesMatch(doc, r.knowledge_title || '') ||
+ docTitlesMatch(doc, r.knowledge_filename || ''),
+ )
+ if (byDoc?.id) return byDoc.id
+ }
+
+ const faqMatch = raw.match(/^FAQ-(\d+)$/i)
+ if (faqMatch) {
+ const faqRefs = list.filter((r) => r.chunk_type === 'faq')
+ const hit = faqRefs[parseInt(faqMatch[1], 10) - 1]
+ if (hit?.id) return hit.id
+ }
+
+ const docMatch = raw.match(/^DOC-(\d+)$/i)
+ if (docMatch) {
+ const docRefs = list.filter((r) => r.chunk_type !== 'faq')
+ const hit = docRefs[parseInt(docMatch[1], 10) - 1]
+ if (hit?.id) return hit.id
+ }
+
+ const num = parseInt(raw, 10)
+ if (!Number.isNaN(num) && String(num) === raw) {
+ const byPos = list[num - 1]
+ if (byPos?.id) return byPos.id
+ const byChunkIndex = list.find((r) => r.chunk_index === num || r.chunk_index === num - 1)
+ if (byChunkIndex?.id) return byChunkIndex.id
+ }
+
+ if (kbId) {
+ const scoped = list.filter((r) => r.knowledge_base_id === kbId)
+ if (doc) {
+ const byDoc = scoped.find(
+ (r) =>
+ docTitlesMatch(doc, r.knowledge_title || '') ||
+ docTitlesMatch(doc, r.knowledge_filename || ''),
+ )
+ if (byDoc?.id) return byDoc.id
+ }
+ if (scoped.length === 1 && scoped[0].id) return scoped[0].id
+ }
+
+ return raw
+}
+
/** Convert / / [[wiki]] tags into inline citation HTML. */
-export function preprocessCitationTags(contentStr: string): string {
+export function preprocessCitationTags(
+ contentStr: string,
+ refs?: CitationKnowledgeRef[] | null,
+): string {
if (!contentStr.trim()) return ''
return contentStr
- .replace(/]*)\/>/g, (_m, attrString: string) => {
+ .replace(WEB_TAG_ATTR_RE, (_m, attrString: string) => {
const attrs = parseTagAttributes(attrString)
const url = attrs.url || ''
const title = attrs.title || ''
@@ -52,20 +144,21 @@ export function preprocessCitationTags(contentStr: string): string {
}
const safeTitle = escapeHtml(title)
const safeUrl = escapeHtml(url)
- return `${domain}${safeTitle}${safeUrl}`
+ return `${domain}${safeTitle}${safeUrl}`
})
- .replace(/]*)\/>/g, (_m, attrString: string) => {
+ .replace(KB_TAG_ATTR_RE, (_m, attrString: string) => {
const attrs = parseTagAttributes(attrString)
const doc = attrs.doc || ''
- const chunkId = attrs.chunk_id || attrs.chunkId || ''
+ const rawChunkId = attrs.chunk_id || attrs.chunkId || ''
const kbId = attrs.kb_id || attrs.kbId || ''
+ const chunkId = resolveCitationChunkId(rawChunkId, { doc, kbId }, refs)
if (!doc || !chunkId) return ''
const safeDoc = escapeHtml(doc)
const safeKbId = escapeHtml(kbId)
const safeChunkId = escapeHtml(chunkId)
const displayDoc = escapeHtml(truncateMiddle(doc))
- return `${displayDoc}…`
+ return `${displayDoc}…`
})
.replace(/\[\[([^\]]+)\]\]/g, (match, inner: string) => {
const pipeIdx = inner.indexOf('|')
@@ -85,7 +178,10 @@ export function preprocessCitationTags(contentStr: string): string {
const HTML_PLACEHOLDER_RE = /@@WEKNORA_HTML_PLACEHOLDER_(\d+)@@/g
/** Protect citation HTML from markdown parser; restore after marked.parse. */
-export function extractCitationHtmlPlaceholders(contentStr: string): { content: string; htmlSnippets: string[] } {
+export function extractCitationHtmlPlaceholders(
+ contentStr: string,
+ refs?: CitationKnowledgeRef[] | null,
+): { content: string; htmlSnippets: string[] } {
const htmlSnippets: string[] = []
const storeHtml = (html: string): string => {
const idx = htmlSnippets.length
@@ -94,8 +190,8 @@ export function extractCitationHtmlPlaceholders(contentStr: string): { content:
}
const content = contentStr
- .replace(/<(?:kb|web)\b[^>]*\/>/g, (match) => storeHtml(preprocessCitationTags(match)))
- .replace(/\[\[([^\]]+)\]\]/g, (match) => storeHtml(preprocessCitationTags(match)))
+ .replace(KB_WEB_TAG_RE, (match) => storeHtml(preprocessCitationTags(match, refs)))
+ .replace(/\[\[([^\]]+)\]\]/g, (match) => storeHtml(preprocessCitationTags(match, refs)))
return { content, htmlSnippets }
}
@@ -105,10 +201,67 @@ export function restoreCitationHtmlPlaceholders(html: string, htmlSnippets: stri
return html.replace(HTML_PLACEHOLDER_RE, (_match, idx) => htmlSnippets[Number(idx)] || '')
}
+/** Collapse newlines around / so marked keeps citations inline. */
+export function joinCitationTagsToPreviousLine(content: string): string {
+ if (!content) return content
+
+ let result = content
+
+ // Newlines between consecutive citation tags
+ let prev = ''
+ while (result !== prev) {
+ prev = result
+ result = result.replace(
+ /(<(?:kb|web)\b[^>]*?\s*\/?>)\s*\n+\s*(<(?:kb|web)\b)/gi,
+ '$1 $2',
+ )
+ }
+
+ // Blank lines before citations: join to previous prose, but keep a break after lists
+ result = result.replace(/\n[ \t]*\n+([ \t]*<(?:kb|web)\b)/gi, (match, kbStart, offset, full) => {
+ const before = full.slice(0, offset)
+ const lastLine = before.split('\n').filter((line) => line.trim()).pop() || ''
+ const isListItem = /^\s*(\d+\.|[-*+])\s+\S/.test(lastLine)
+ return isListItem ? `\n\n${kbStart}` : ` ${kbStart}`
+ })
+
+ // Single newline before citation when it follows text or another citation (not after a blank line)
+ result = result.replace(
+ /(?]*?\s*\/?>|[ \t]*\S[^\n]*?)\n([ \t]*<(?:kb|web)\b)/g,
+ '$1 $2',
+ )
+
+ return result
+}
+
+const CITATION_HTML_FRAGMENT =
+ '(?:)'
+
+/** Merge citation-only blocks into the preceding paragraph (marked splits on newlines). */
+export function collapseStandaloneCitationParagraphs(html: string): string {
+ if (!html || !html.includes('citation')) return html
+
+ const mergePattern = new RegExp(
+ `(<\\/(?:p|li)>)\\s*(?:
\\s*<\\/p>\\s*)*
\\s*(${CITATION_HTML_FRAGMENT})\\s*<\\/p>`,
+ 'g',
+ )
+
+ let result = html
+ let prev = ''
+ while (result !== prev) {
+ prev = result
+ result = result.replace(mergePattern, (_match, closeTag: string, citation: string) => {
+ return ` ${citation}${closeTag}`
+ })
+ }
+
+ return result
+}
+
/** Preserve raw / tags before sanitizers that would strip them. */
export function preserveCitationTags(contentStr: string): { text: string; tags: string[] } {
const tags: string[] = []
- const text = contentStr.replace(/<(?:kb|web)\b[^>]*\/>/g, (match) => {
+ const text = contentStr.replace(KB_WEB_TAG_RE, (match) => {
const idx = tags.length
tags.push(match)
return `\x00TAG${idx}\x00`
diff --git a/frontend/src/views/chat/components/botmsg.vue b/frontend/src/views/chat/components/botmsg.vue
index 2092a6c2..7bc9bf94 100644
--- a/frontend/src/views/chat/components/botmsg.vue
+++ b/frontend/src/views/chat/components/botmsg.vue
@@ -3,23 +3,27 @@
-
+
-
+
{{ item.name }}
-
-
+
+
+
+
+
@@ -40,10 +44,12 @@