An assistant that invents answers is worse than no assistant at all. Grounding is the whole game.
Large language models are fluent, which is exactly the problem: they will confidently produce something plausible even when they do not know. For support, that is unacceptable.
Retrieval first, generation second
Before the model writes anything, KuraChat retrieves the most relevant passages from your own content — your docs, FAQs, and uploaded files — and asks the model to answer strictly from them.
Scoped to your workspace
Retrieval is scoped to a single workspace. One customer’s content can never leak into another’s answers, and the assistant will not reach outside what you gave it.
If the answer is not in your content, the right response is “I am not sure, let me get a person” — not a guess.
Confidence and escalation
Every answer carries a confidence signal. Low confidence can trigger a handoff automatically, so uncertain questions reach your team instead of getting a shaky reply.