feat(knowledge): search results a person can open, from Chat

- The composer's Search mode searches every knowledge base as the signed-in
  person and lists what they may read as result cards: source icon, title
  linking back to the document, knowledge base, updated date, and the
  matching passage with the query terms in bold; Summarize hands a document
  to the agent in Build mode
- The agent's knowledge tool returns each result's title, link, connector,
  and modified time and is told to cite with source tags carrying a snippet;
  a reply whose sources carry snippets ends with the same cards
- A session route for the search, bound to the shared search use case, so
  the browser reads through the same access predicate as everything else
- Search quality: hybrid legs over-fetch before fusion, the vector leg's
  iterative scan fills a limit past the default candidate pool, and the
  recency weight moves a fresh document a few places rather than the list
- A manual member sync makes every active member due, so Sync members now
  lists everyone instead of nobody

Author: waleedlatif1

apps/sim/app/api/knowledge/search/route.ts

@@ -0,0 +1,47 @@
+import { searchWorkspaceKnowledgeContract } from '@/lib/api/contracts/knowledge'
+import {
+  defineInternalJsonRoute,
+  internalRateLimits,
+  internalSessionAuth,
+} from '@/lib/api/server/routes'
+import { internalKnowledgeErrorPolicies } from '@/lib/knowledge/api/route-policies'
+import { knowledgeOperations } from '@/lib/knowledge/application/operations'
+import { searchKnowledge } from '@/lib/knowledge/application/search'
+
+export const POST = defineInternalJsonRoute({
+  contract: searchWorkspaceKnowledgeContract,
+  auth: internalSessionAuth,
+  operation: knowledgeOperations.search,
+  rateLimit: internalRateLimits.none({
+    reason: 'A person typing queries; the embedding call is metered against their workspace',
+  }),
+  errorPolicy: internalKnowledgeErrorPolicies.search,
+  mapInput: ({ body }) => ({
+    workspaceId: body.workspaceId,
+    knowledgeBaseIds: body.knowledgeBaseIds,
+    query: body.query,
+    topK: body.topK,
+  }),
+  useCase: searchKnowledge,
+  present: ({ results, knowledgeBases }, { input }) => {
+    const knowledgeBaseNames = new Map(knowledgeBases.map((kb) => [kb.id, kb.name]))
+    return {
+      success: true as const,
+      data: {
+        query: input.query ?? '',
+        results: results.map((result) => ({
+          documentId: result.documentId,
+          knowledgeBaseId: result.knowledgeBaseId,
+          knowledgeBaseName: knowledgeBaseNames.get(result.knowledgeBaseId) ?? '',
+          documentName: result.documentName,
+          sourceUrl: result.sourceUrl,
+          connectorType: result.connectorType,
+          sourceModifiedAt: result.sourceModifiedAt?.toISOString() ?? null,
+          content: result.content,
+          chunkIndex: result.chunkIndex,
+          similarity: result.similarity,
+        })),
+      },
+    }
+  },
+})

apps/sim/app/api/knowledge/search/utils.test.ts

@@ -510,12 +510,13 @@ describe('Knowledge Search Utils', () => {
 
     it('runs both legs and fuses them in hybrid mode', async () => {
       /**
-       * Chains dequeue in creation order. A workspace-scoped vector leg selects
-       * directly (the iterative scan is reserved for a personal token set), so
-       * its select is built first, then the keyword ranking pass, then hydration.
+       * Chains dequeue in creation order. Hybrid legs over-fetch past the
+       * plain scan's candidate pool, so the vector leg opens its transaction
+       * and applies the scan settings before selecting: the keyword ranking
+       * pass is built first, then the vector select, then hydration.
        */
-      queueTableRows(schemaMock.embedding, [makeResult('vector-hit')])
       queueTableRows(schemaMock.embedding, [{ id: 'keyword-hit', keywordRank: 0.9 }])
+      queueTableRows(schemaMock.embedding, [makeResult('vector-hit')])
       queueTableRows(schemaMock.embedding, [makeResult('keyword-hit')])
 
       const results = await executeKnowledgeSearch({
@@ -532,9 +533,9 @@ describe('Knowledge Search Utils', () => {
     })
 
     it('falls back to vector results when the keyword leg fails', async () => {
-      queueTableRows(schemaMock.embedding, [makeResult('vector-hit')])
-      /** The failing ranking chain is still built and takes the second queued set. */
+      /** The failing ranking chain is still built first and takes the first queued set. */
       queueTableRows(schemaMock.embedding, [{ id: 'never-ranked', keywordRank: 0 }])
+      queueTableRows(schemaMock.embedding, [makeResult('vector-hit')])
 
       /**
        * Both legs share one `orderBy` spy, so target the keyword leg by its

apps/sim/app/workspace/[workspaceId]/home/components/knowledge-search-results/index.ts

@@ -0,0 +1 @@
+export { groupResultsByDocument, KnowledgeSearchResults } from './knowledge-search-results'

apps/sim/app/workspace/[workspaceId]/home/components/knowledge-search-results/knowledge-search-results.tsx

@@ -0,0 +1,129 @@
+'use client'
+
+import { useMemo } from 'react'
+import type { WorkspaceKnowledgeSearchResult } from '@/lib/api/contracts/knowledge'
+import { SourceCard } from '@/app/workspace/[workspaceId]/home/components/message-content/components/source-card'
+import type { SourceTagData } from '@/app/workspace/[workspaceId]/home/components/message-content/components/special-tags'
+import { useKnowledgeBasesQuery, useWorkspaceKnowledgeSearch } from '@/hooks/queries/kb/knowledge'
+
+/** A search spans at most this many knowledge bases. */
+const MAX_SEARCHED_KNOWLEDGE_BASES = 20
+/** Characters of the matching chunk shown under a result. */
+const SNIPPET_LENGTH = 280
+
+function toSnippet(content: string): string {
+  const flat = content.replace(/\s+/g, ' ').trim()
+  return flat.length > SNIPPET_LENGTH ? `${flat.slice(0, SNIPPET_LENGTH).trimEnd()}…` : flat
+}
+
+/**
+ * One card per document, keeping the best-ranked chunk of each: the list is
+ * already in rank order, so the first chunk seen for a document is its best.
+ */
+export function groupResultsByDocument(
+  results: readonly WorkspaceKnowledgeSearchResult[]
+): WorkspaceKnowledgeSearchResult[] {
+  const seen = new Set<string>()
+  const grouped: WorkspaceKnowledgeSearchResult[] = []
+  for (const result of results) {
+    if (seen.has(result.documentId)) continue
+    seen.add(result.documentId)
+    grouped.push(result)
+  }
+  return grouped
+}
+
+/** A result as the source card renders it; a document without a source URL cannot be opened. */
+function toSource(result: WorkspaceKnowledgeSearchResult): SourceTagData | null {
+  if (!result.sourceUrl) return null
+  return {
+    url: result.sourceUrl,
+    title: result.documentName ?? undefined,
+    siteName: result.knowledgeBaseName || undefined,
+    connectorType: result.connectorType ?? undefined,
+    snippet: toSnippet(result.content),
+    updatedAt: result.sourceModifiedAt ?? undefined,
+  }
+}
+
+interface KnowledgeSearchResultsProps {
+  workspaceId: string
+  query: string
+  /** Asks the agent about one document; the prompt names it and links to it. */
+  onSummarize: (prompt: string) => void
+}
+
+/**
+ * The composer's Search mode: the documents the signed-in person may read that
+ * match their query, across every knowledge base in the workspace, as cards
+ * that open the source. Summarize hands one document to the agent.
+ */
+export function KnowledgeSearchResults({
+  workspaceId,
+  query,
+  onSummarize,
+}: KnowledgeSearchResultsProps) {
+  const { data: knowledgeBases = [], isPending: basesPending } = useKnowledgeBasesQuery(workspaceId)
+  const knowledgeBaseIds = useMemo(
+    () => knowledgeBases.slice(0, MAX_SEARCHED_KNOWLEDGE_BASES).map((kb) => kb.id),
+    [knowledgeBases]
+  )
+  const {
+    data: results,
+    isPending,
+    isFetching,
+    error,
+  } = useWorkspaceKnowledgeSearch(workspaceId, knowledgeBaseIds, query)
+  const documents = useMemo(() => groupResultsByDocument(results ?? []), [results])
+
+  if (!basesPending && knowledgeBaseIds.length === 0) {
+    return (
+      <p className='px-2 py-3 text-[var(--text-muted)] text-small'>
+        No knowledge bases to search yet. Add one from the Knowledge tab.
+      </p>
+    )
+  }
+  if (error) {
+    return <p className='px-2 py-3 text-[var(--text-error)] text-small'>{error.message}</p>
+  }
+  if (isPending || (isFetching && !results)) {
+    return <p className='px-2 py-3 text-[var(--text-muted)] text-small'>Searching…</p>
+  }
+  if (documents.length === 0) {
+    return (
+      <p className='px-2 py-3 text-[var(--text-muted)] text-small'>
+        No documents you can read match “{query}”.
+      </p>
+    )
+  }
+
+  return (
+    <div className='flex flex-col gap-0.5'>
+      {documents.map((result) => {
+        const source = toSource(result)
+        return source ? (
+          <SourceCard
+            key={result.documentId}
+            source={source}
+            query={query}
+            onSummarize={(cited) =>
+              onSummarize(`Summarize "${cited.title ?? cited.url}" (${cited.url})`)
+            }
+          />
+        ) : (
+          <div key={result.documentId} className='flex flex-col gap-0.5 px-2 py-2'>
+            <p className='truncate text-[var(--text-primary)] text-sm'>
+              {result.documentName ?? 'Untitled document'}
+            </p>
+            <p className='truncate text-[var(--text-muted)] text-caption'>
+              {result.knowledgeBaseName}
+            </p>
+            <p className='line-clamp-2 text-[var(--text-body)] text-small leading-snug'>
+              {toSnippet(result.content)}
+            </p>
+          </div>
+        )
+      })}
+    </div>
+  )
+}

apps/sim/app/workspace/[workspaceId]/home/components/message-content/components/index.ts

@@ -4,5 +4,6 @@ export { ChatContent } from './chat-content'
 export { MessageSources } from './message-sources'
 export { Options } from './options'
 export { QuestionDisplay } from './question'
+export { highlightTerms, SourceCard } from './source-card'
 export { SourceChip, sourceLabel } from './source-chip'
 export { PendingTagIndicator, parseSpecialTags, SpecialTags } from './special-tags'

apps/sim/app/workspace/[workspaceId]/home/components/message-content/components/message-sources/message-sources.tsx

@@ -1,6 +1,7 @@
 'use client'
 
 import { cn } from '@sim/emcn'
+import { SourceCard } from '@/app/workspace/[workspaceId]/home/components/message-content/components/source-card'
 import { SourceChip } from '@/app/workspace/[workspaceId]/home/components/message-content/components/source-chip'
 import type { SourceTagData } from '@/app/workspace/[workspaceId]/home/components/message-content/components/special-tags'
 
@@ -14,17 +15,41 @@ const STRIP_FADE_CLASSES =
 
 interface MessageSourcesProps {
   sources: readonly SourceTagData[]
+  /** The question the reply answers; its terms are bolded in result cards. */
+  query?: string
+  /** Asks the agent about one cited document, when the surface can send a message. */
+  onSummarize?: (prompt: string) => void
 }
 
 /**
- * Footer strip listing every document a reply cited, once each: one
- * horizontally scrolling row of {@link SourceChip}s that fades out at the right
- * edge instead of wrapping, so a long list stays a single quiet line under the
- * answer.
+ * Footer listing every document a reply cited, once each. Sources that carry
+ * a snippet — a search answer — are laid out as result cards; otherwise one
+ * horizontally scrolling row of {@link SourceChip}s that fades out at the
+ * right edge instead of wrapping, so a long list stays a single quiet line
+ * under the answer.
  */
-export function MessageSources({ sources }: MessageSourcesProps) {
+export function MessageSources({ sources, query, onSummarize }: MessageSourcesProps) {
   if (sources.length === 0) return null
 
+  if (sources.some((source) => source.snippet)) {
+    return (
+      <div className='flex flex-col gap-0.5'>
+        {sources.map((source) => (
+          <SourceCard
+            key={source.url}
+            source={source}
+            query={query}
+            onSummarize={
+              onSummarize
+                ? (cited) => onSummarize(`Summarize "${cited.title ?? cited.url}" (${cited.url})`)
+                : undefined
+            }
+          />
+        ))}
+      </div>
+    )
+  }
+
   return (
     <div
       className={cn(

apps/sim/app/workspace/[workspaceId]/home/components/message-content/components/source-card/index.ts

@@ -0,0 +1 @@
+export { highlightTerms, SourceCard } from './source-card'