Block

Source-attributed answer

RAG-style answer with inline citations. Each chip is hover-previewable; click to open the source. Numbered footer summarises everything cited.

Inline cites + sources footer

What's the difference between HNSW and IVF vector indexes?

HNSW (Hierarchical Navigable Small World) builds a layered graph where each layer skips long-distance connections, so search descends through layers narrowing the candidate set 1 . IVF (Inverted File) clusters vectors with k-means and only searches a handful of nearby clusters at query time 2 . HNSW typically gets higher recall at the same latency, but IVF uses less memory and is easier to update incrementally 3 .

Sources

wikipedia.org

pinecone.io

github.com

<div hlmMessage>
  <pk-message-avatar src="" alt="Assistant" fallback="AI" />
  <div class="flex flex-1 flex-col gap-3">
    <p class="text-sm leading-relaxed">
      HNSW builds a layered graph
      <pk-source href="https://en.wikipedia.org/wiki/Hierarchical_navigable_small_world">
        <pk-source-trigger label="1" />
        <pk-source-content
          title="HNSW — Wikipedia"
          description="A graph-based ANN structure with O(log N) query complexity."
        />
      </pk-source>
      . IVF clusters vectors
      <pk-source href="https://www.pinecone.io/learn/series/faiss/vector-indexes/">
        <pk-source-trigger label="2" />
        <pk-source-content title="Vector indexes" description="..." />
      </pk-source>
      .
    </p>

    <div class="border-t pt-3">
      <span class="text-xs uppercase">Sources</span>
      <div class="flex flex-wrap gap-2">
        <pk-source href="https://en.wikipedia.org/wiki/...">
          <pk-source-trigger [showFavicon]="true" label="wikipedia.org" />
          <pk-source-content title="..." description="..." />
        </pk-source>
        <!-- ...more sources... -->
      </div>
    </div>
  </div>
</div>