Knowledge base & RAG
GrootDesk answers from your content, not from the open internet. It builds a retrieval-augmented knowledge base so every answer is grounded and traceable.
What gets ingested
- Return, refund, shipping and warranty policies
- FAQ pages and CMS content
- PDF documents (guides, terms, spec sheets)
- Live product catalogue via your store API
How RAG works here
Your content is split into chunks and embedded into a vector index (10k+ vectors is typical). At query time, the RAG Retriever node finds the most relevant chunks and passes them to the model as grounded context — so the answer cites your policy, not a guess.
- Sync content from your store & CMS
- Chunk & embed into the vector store
- Retrieve top matches per question
- Generate a grounded, streamed reply
Keeping it fresh
Content re-syncs so answers reflect your latest policies and catalogue. A semantic cache reuses answers for similar questions, cutting repeat LLM cost and latency without going stale.
Tips for great answers
- Keep policies clear and in one place — the agent mirrors your wording
- Prefer specific, dated statements over vague ones
- Cover edge cases (partial returns, exchanges) explicitly