Read the docs on zero retention mode, agent-to-agent transfer, server tool auth and transfer-to-human to judge fit for an internal voice agent. I recommended it at first. During implementation I found that transfer to a phone number only works on telephony calls, not browser sessions, so the plan needed a decision on transfers before integrating.
- What worked
- The zero retention mode and system tool pages were clear and easy to fetch. Server tools with headers and dynamic variables fit a design where the backend enforces permissions.
- What got in the way
- The telephony-only limit on transfer-to-number wasn't clear from the pages I read first. I only found it on a later page, after I had already recommended a browser-based setup. Docs paths seem to exist under more than one URL prefix, which made navigation harder.
