Answers with receipts
Every AI answer can now cite the exact knowledge it came from — and a rebuilt retrieval pipeline with reranking and a groundedness check makes those answers measurably more accurate.
Trust in an AI agent comes down to one question: where did that answer come from? Answers now cite their sources — the exact articles, pages, or documents the agent drew from — right in the conversation.
Behind it sits a rebuilt retrieval pipeline: semantic and keyword search run together, a reranking stage promotes the passages that actually answer the question, and an automated groundedness judge scores every answer against the knowledge it cited. The result in our evals: substantially more accurate answers, and a paper trail for every one of them.
Grounded by design
The agent answers from your knowledge base — your website, your articles, your uploaded files — not from general internet knowledge. If something isn't in the knowledge base, the agent won't claim to know it. Every reply follows the same pipeline: retrieve the relevant material, optionally call a tool, then write the answer from what was found — recording which knowledge base items it drew on.
Retrieval happens per question, so the agent always works from the latest indexed version of your content. There is no separate “publish to the agent” step: articles re-index the moment you save, documents at upload, websites on sync.
Inspect every answer in the Playground
The Playground's Response details rail shows what the agent did behind the scenes for its latest reply: the sources cited, every tool call with expandable arguments and results, a thumbs up/down to rate the answer, and a snapshot of the live configuration it ran with. Ask a question, check which passages it drew on, adjust your knowledge, re-run.
See what's new in your workspace
Everything on this page is live today. Asks trains on your website and resolves customer conversations on every channel — free to try, live in minutes.