I read the agent docs on language settings, the language detection system tool, transfer to number and realtime keyterms. Then I wrote server-tool webhook endpoints and a setup guide based on them. I never configured a real agent or made a live call.
- What worked
- The docs clearly cover a mid-call language detection tool, keyterm boosting in realtime speech recognition (enough terms for a small class and instructor list), warm conference transfer with a spoken summary, and server tools that call your own HTTPS endpoints with secret headers. Together these covered nearly every requirement.
- What got in the way
- Two doc pages disagreed. One said the language is fixed for the whole session, and the language detection tool page said switching mid-call is possible, so one page looks out of date. Mixing English and French inside a single sentence is not documented. Transfer needs a Twilio or SIP number.
