Customer support RAG over your knowledge base Human in the loop

A support agent that refuses to make things up.

A customer asks. It answers from your help center and shows the exact article it used. Anything it cannot source goes to a person, not a guess.

The same brain runs on your website, WhatsApp or Slack, and answers in the language your customer wrote in.

1

A question comes in

From your widget, WhatsApp or Slack.

2

Articles retrieved

The closest help articles, ranked by the query.

3

Answered with a source

The exact article and the line it used.

4

Confident or escalated

Refunds and cancellations get a ticket and a person.

Nothing to cite means it refuses

No matching article, so a human takes over. It does not invent an answer.

The problem

A wrong answer about a refund costs real money.

Customers ask about refunds, prices and cancellation terms. Answer one of those with a guess and you are on the hook for a promise you never made.

So the answer has to come from your real help content, and the bot has to know when to stop and pass the question to a person.

This build answers from your knowledge base, cites the article it used, and sends anything sensitive to a person. Fast for the customer, and every answer traces back to a source you control.

  • Every answer is sourced. It names the exact help article and highlights the line it used.
  • Sensitive cases escalate. Refunds and cancellations go to a person, and the rule is enforced in code.
  • Unknowns are refused. No matching article means a ticket for a human, not a guess.
Try it live

Push a real question through it.

Shown here on a demo help center. Tap a saved example, or type your own trap. Simulated data, real engineering.

basaltflow.ai/support Live

This runs the same engine I would build for you. Try to make it answer something it cannot source.

Want this on your knowledge base?

Tell me where your help content lives and what your customers keep asking. I will reply with the fastest honest way to wire it up.