Skip to main content
aifeatures.ai
Knowledge teams chunking PDFs into Pinecone, not a Supabase row

The OpenRouter chunk stack

Make splits the corpus, OpenRouter embeds, Pinecone stores, Notion lists what landed. Postgres stays infra.

Who owns what

  1. Make

    The crawl — file in, upsert out. Not embeddings.

    Try Make (opens in a new tab)

  2. OpenRouter

    The chunk — embeddings and section titles. Not the index.

    Try OpenRouter (opens in a new tab)

  3. Pinecone

    The vectors — namespace, ids. Not the inventory page.

    Try Pinecone (opens in a new tab)

  4. Notion AI

    The inventory — source, chunk count. Not cosine.

    Try Notion AI (opens in a new tab)

How to use this stack

  1. Drop the PDF set on Make. Do not stand up Supabase as a fourth personality.
  2. OpenRouter returns chunks plus embeddings. One router — not Bedrock on the same corpus.
  3. Make upserts Pinecone, then writes the inventory row in Notion.
  4. This is not enterprise-rag-router (that one answers Slack). Not the Bedrock SOP (PHI Guardrails).