Last updated November 13, 2026.
Grounding
The agent answers only from the content a business gives it: its website, documents and approved answers. Each answer links its source. Numbers such as prices may appear only when a cited source contains them. When the content has no answer, the agent says so.
Disclosure
The widget shows that customers are talking to an AI assistant. The agent confirms it is an AI when asked and never claims to be a person.
Human handoff
People stay in charge. The agent hands a conversation to the business’s team on low confidence, on request, on complaints and in emergencies, and a team member can join any conversation at any time. Businesses can require approval before replies are sent.
Learning under supervision
Nothing the agent learns from conversations becomes knowledge until a person approves it. Personal details are removed before a question is proposed.
Safety
Industry templates include emergency phrases in each supported language. When one appears, the customer gets a safety message and the business’s on-call person is alerted. The agent does not give medical, legal or financial advice.
Protection against manipulation
Text on crawled pages that tries to instruct the agent is flagged and ignored. Messages that try to change the agent’s instructions are screened.
Measurement
Businesses see the numbers behind the agent’s work with formulas. A sample of conversations is scored for grounding, resolution, tone and policy so problems surface quickly.
Contact
Questions about how the agent works: the address on our contact page (/contact)