Solutions 09 · RAG Q&A over Your Docs

Exercise 1

On Add to Knowledge Base, set Clear Store to the expression {{ $('Ingest API').item.json.body.replace === true }}.

jq -n --rawfile t assets/sample-handbook.md '{text: $t, source: "handbook", replace: true}' \
  | curl -X POST http://localhost:5678/webhook/kb/ingest -H 'Content-Type: application/json' -d @-

Clearing wipes everything under that key. For multi-document knowledge bases, use a persistent store with a delete-by-metadata step (filter source) instead.

Exercise 2

Qdrant (self-hosted): add to playground/docker-compose.yml:

  qdrant:
    image: qdrant/qdrant
    volumes:
      - qdrant_data:/qdrant/storage

(and qdrant_data: under volumes:). Create a Qdrant credential in n8n with URL http://qdrant:6333. Replace both Simple Vector Store nodes with Qdrant Vector Store (collection course09_kb): insert mode for ingest, Retrieve Documents (As Tool) for search_docs. The embedding sub-nodes stay the same.

Supabase: create a project, run the pgvector setup SQL from n8n's Supabase Vector Store docs (a documents table + match_documents function), add the Supabase credential, and swap the nodes the same way.

Exercise 3

1. Change Has Text? to also accept a PDF: {{ String($json.body?.text ?? '').trim().length >= 20 || Object.values($binary ?? {}).some(f => f.mimeType === 'application/pdf') }} (Boolean is true).
2. Set the Default Data Loader's type with an expression, or simpler: add a second insert branch for PDFs with a loader of type Binary, loader PDF Loader, input field taken from the upload (file).
3. Set metadata source = {{ Object.values($binary ?? {})[0]?.fileName ?? $json.body?.source ?? 'unnamed' }}.

curl -X POST http://localhost:5678/webhook/kb/ingest -F "file=@policy.pdf;type=application/pdf"