●  RETRIEVAL-AUGMENTED · CITES EVERY ANSWER · HUMAN-IN-THE-LOOP

Grounded answers.
Real actions.
An honest handoff.

RAG Support Agent is a drop-in support widget. Every reply is grounded in your help docs with citations you can click. It calls tools to look up live account data — and when it isn't sure, it hands off to a human instead of guessing.

Try the live demo How it works
Measured · RAGAS-style eval
Faithfulness0.95
Deflection (no handoff)78.7%
Out-of-scope escalation100%
Multi-doc completeness@5 (graph on)53→63%
Cost / question$0.0093
Every answer is grounded and cited — or the agent escalates. Quality is measured, not asserted.
01 Grounded in your docs
02 Inline, clickable citations
03 Live tool calls
04 Human handoff
05 Measured quality
06 Multi-hop retrieval
Live · support agent

Ask it anything.
It answers with receipts.

This demo is grounded in a 19-document help-doc set and a mock order system. Every answer cites its source. For account questions it looks things up live. When a question spans several documents, a knowledge-graph retrieval arm connects them. When it can't help, a human takes over.

Try the chat in the corner →
●  Authenticated access · role-based ACL

Most answers are public. Some procedures live in internal-only docs. Sign in as an agent to unlock them — an anonymous visitor asking the same thing gets an honest handoff instead.

01 · grounded
Grounded, with citations
Answers are retrieved from your docs and cited inline. Click a citation to read the source passage. Below the confidence bar, the model gets no context — so it can't invent one.
02 · agentic
Takes real actions
It's a real tool-use loop, not a single pass. The agent looks up live order and account data mid-conversation, and shows exactly which tool it ran and what came back.
03 · human-in-the-loop
Knows when to stop
Low confidence or out of scope? It escalates to a human teammate instead of guessing — with a calm handoff, not an error. Escalations are logged and measured.