Every answer from your docs shows its work
Trust in AI answers comes from provenance. Every Kelu answer from your docs carries numbered citations to the exact source documents used — and when the knowledge isn't there, the assistant says so instead of inventing something.
Grounding, end to end
From retrieval to the moment a user clicks a source
Source-Backed Answers
- Numbered sources assembled into the answer context
- Structured citations array: URL, title, excerpt
- Click through to the exact source page
- Citations across widget, bots, portal, API, and MCP
- Related articles surfaced alongside answers
Honest Uncertainty
- Confidence scoring on every answer
- Low-confidence fallback instead of hallucination
- Configurable fallback routes users to support
- Unanswerable questions logged as knowledge gaps
- Form deflection only deflects well-grounded answers
Verifiability
- Conversation history with full citation records
- Traces show each question and answer, with latency and tokens
- User feedback (helpful / not helpful) per message
- Citation coverage shown in analytics
- Audit-friendly for compliance review
The difference between an answer and a liability
A support answer that invents a refund policy, a fee schedule, or an API parameter costs more than no answer at all. Kelu generates answers only from content you indexed and approved, cites it, and falls back transparently outside that boundary — which is what makes it deployable in fintech, healthcare, and the public sector.
- Answers restricted to your indexed, approved content
- Users verify claims with one click on the citation
- Compliance can review any stored conversation with its citations
- Deflection metrics count only confident, grounded answers
- Gaps become a prioritized docs backlog, not silent failures
Yes, within 30 days of purchase1. The refund goes back to the card you paid with2.
Sources
https://docs.acme.dev/billing/refunds
The Acme docs don’t cover HIPAA agreements, so I can’t answer that from them.
Answers you can check, from a system that won't guess
Every answer from your docs shows its sources
Numbered links point to the page and section each part of the answer came from, so anyone can check it in one click.
It says "I don't know"
When your content does not cover a question, Kelu does not guess. In a helpdesk it posts nothing and tags the ticket for your team.
A bad sync can't empty it
If one sync would remove more than 45% of a source's pages, Kelu holds the removals until a person approves them.
Got questions?
Frequently Asked Questions
Common questions about Citations & Grounding
Each citation includes the source URL, document title, and the excerpt used, delivered as a structured array alongside the answer — so any UI, bot, or API consumer can render clickable, verifiable sources.
Answers are generated from retrieved chunks of your indexed content, the prompt requires grounding in those sources, and confidence scoring gates the result. When your sources don't cover the question, the excerpts and citations are withheld and users are told it can't answer — not a guess.
Yes, unless you turned on zero data retention or set a shorter retention window. Conversations are stored with their citations, traces record each question and answer with its latency and tokens, and workspace audit logs cover administrative actions.
Yes — widget, Slack and Microsoft Teams bots, the Ask portal, the public API, and MCP responses all carry the same structured citations from the same pipeline.
They become data. Gap analytics collect unanswered and low-confidence questions into a prioritized list, and Knowledge Maintenance drafts answers for your team to review and publish.
Ship AI answers you can defend
Grounded, cited, and auditable — try it on your own docs for free.
Free to start · No credit card required