Modern AI voice agents combine three components: a real-time speech-to-text engine, an LLM (often a tuned variant of GPT or Claude) for response generation, and a high-quality text-to-speech engine. End-to-end latency is typically 600–1,200ms.
Callsy agents disclose their AI status at call commencement per EU AI Act and US TCPA expectations. Disclosed AI calls perform as well as undisclosed in conversion tests.
Use cases: abandoned cart recovery, inbound lead qualification, appointment booking, win-back, payment reminders, NPS surveys.