Demo A · RAG Knowledge Chatbot

DocsAnswer

A chatbot that answers from your documentation — with a source citation on every claim, and a guardrail that makes it say "I don't know" instead of guessing. Pick a question; the answer's citations light up the exact doc passage they came from.

Demo mode: answers are pre-computed and replayed in your browser — no API key, no cost. The live version connects to the Claude API over a real vector store (embed → retrieve → cite-grounded generation). The knowledge base below is 15 synthetic help pages for a fictional SaaS, "Lumina Notes".

Try a question — the last two are out-of-scope on purpose

Select a question to see a grounded, cited answer.

Knowledge base · 15 pages cited passages highlight →

How the answer is grounded

01 · INGEST

Chunk & embed

Docs are split into small passages and embedded into a vector store.

02 · RETRIEVE

Find top passages

The question is embedded; the closest passages are pulled back.

03 · GENERATE

Answer, cite-forced

Claude answers using only those passages and must cite each claim.

04 · GUARD

Refuse if unsupported

If the passages don't answer it, the model says so and defers to a human.

Why this matters: the #1 objection to support chatbots is that they hallucinate confident wrong answers. Grounding every sentence in a retrievable, clickable source — and refusing when the source isn't there — is what makes a RAG bot safe to put in front of your customers.