RAG Q&A over Your Docs

After this module you can build an assistant that answers questions from your documents, with sources, and says "I don't know" instead of making things up.

What it does

Two flows in one workflow:

• Ingest (POST /webhook/kb/ingest {text, source}): splits the document into ~800-character chunks, embeds them with OpenAI text-embedding-3-small, and stores them in n8n's in-memory vector store.
• Ask (hosted chat): an AI Agent searches the knowledge base with a search_docs tool, answers only from what it finds, and cites the source.

A sample company handbook (assets/sample-handbook.md) with known answers is included for testing.

Flow

Ingest API → Has Text? ─true─→ Add to Knowledge Base (vector store: insert) → Respond Ingested
                       └false→ 400        ├ Embeddings (ingest)
                                          └ Document Loader ← Text Splitter

Ask the Docs → Docs Agent ─ model: OpenAI · memory: 6 messages · tool: search_docs (vector store: retrieve) ← Embeddings (search)

Verification status

• Ingest validation (400) and chat page: tested on n8n 2.43.6. A valid ingest reaches the vector store and stops at "Error in sub-node Embeddings (ingest)" without a credential, as expected.
• Embedding, retrieval and answer quality: needs an OpenAI credential. WITH_CREDS=1 ./test/test.sh ingests the handbook and checks 6 answers, including an "I couldn't find that" case.

Files

File	Tier	Purpose
lesson.md	Free	Step-by-step build + how RAG works
workflow/starter.json	Free	Triggers and validation only
workflow/learner.json	Learner	Finished workflow
workflow/business.json	Business	Business workflow (see setup guide)
deploy/docker-compose.qdrant.yml	Business	Qdrant for the playground
setup-guide.md	Business	Credentials + configuration
assets/sample-handbook.md	Learner	Test document with known answers
exercises.md	Learner	Practice tasks
solutions.md	Learner	Exercise solutions
notes.md	Private	Learning log
sell/	Private	Offer, video script, blog post