Every answer traces back to something you wrote.
Grounding connects the agent to your actual rules: return windows, price books, proration methods, clinic calendars. The agent answers from those sources and shows which one it used. If no source covers the question, it doesn't make one up.
Connected sources
- returns-policy.pdfDocument Synced
- Help centre, 42 articlesKnowledge base Synced
- Plans and price bookStructured rules Synced
- Clinic calendarsLive records Synced
An answer, with its source
Can I return the sale jacket?
Sale items can be exchanged within 15 days, but not refunded. I can start an exchange if you'd like.
Source
returns-policy.pdf, section 4.2, version synced 2 days ago
“Items bought at a discount of 30% or more are eligible for exchange within 15 days of delivery and are not eligible for refund.”
What you can connect
Documents, rules and live records.
Most teams start with the policy documents they already have, then add structured rules and live records for the intents they want the agent to act on.
Policy documents
PDF, Word, Markdown or plain text. Headings and sections are kept, so citations point to the exact clause.
Help-centre articles
Your public or internal knowledge base, imported and kept in step as articles change.
Structured rules
Windows, price books, proration methods, visit types and buffers, entered as rules rather than prose so figures are exact.
Live records
Orders, subscriptions and bookings read through an API at the moment of the conversation, never from a stale copy.
How answers stay grounded
No source, no answer.
The agent is built to prefer a clean handoff over a confident guess. Grounding is enforced on every reply, not checked afterwards.
Cited replies
Each answer is linked to the source section it relied on, visible to your team in the conversation view.
Refusal to invent
When nothing in your sources covers a question, the agent says it will check with the team and hands off.
Conflict detection
If two sources disagree, for example an old article and a new policy, the agent flags the conflict instead of picking one.
Version history
You can see which version of a policy was active when a past answer was given.
Keeping it current
Change the policy once. The agent follows.
Policies change: a festive return window, a new plan, a clinician on leave. Updates flow to the agent without retraining.
Re-sync on change
Connected documents and articles are re-read when they change, and the new version takes effect for new conversations.
Scheduled rules
Set a rule to start and end on specific dates, such as an extended return window for a sale period.
Gap reports
Questions that arrived with no matching source are listed, so you can see which policies are missing.
Review before publish
Optional approval step so a changed rule is checked by your team before the agent starts using it.
Questions
Policy & Knowledge Grounding
Do our documents need a special format?
No. Clear headings help citations point to the right clause, but the agent works with the documents you already use.
What if our policy is only in people's heads?
Then writing it down is the first step, and we'll help structure it during onboarding. The agent can't be more consistent than the rules it is given.
Can different teams have different rules?
Yes. Rules can be scoped by brand, region, product line or clinic location, and the agent applies the one that matches the conversation.
Which question fills your inbox?
Pick the one that sounds most like your customers. We'll set you up with that agent first.