Twilio ConversationRelay

ConversationRelay connects phone calls to AI. When a call arrives on a Twilio number, the service converts the caller’s speech to text in real time and streams it over a WebSocket to the developer’s server. The server sends text back, which Twilio converts into speech, so the caller hears a natural response. This lets teams plug in any large language model and their own business logic.

Twilio manages the hard parts of voice. It handles interruptions when a caller speaks over the agent, supports multiple languages and voice providers, and can transfer the call to a human with the transcript. It also offers a no-code Studio widget for simpler flows, plus compliance support for regulated industries, including options for HIPAA and PCI. A reporting feature gives insight into latency and where agents struggle.

ConversationRelay is billed by usage on top of regular Twilio voice costs, and speech and model providers add their own charges. Latency matters a lot in phone conversations, so design and testing take effort. Alternatives include Vapi, Retell AI, Bland AI, LiveKit, and Amazon Connect. Twilio ConversationRelay suits developers building AI phone agents.