Retell AI is a platform for AI voice agents: programs that hold a spoken phone conversation, answer questions and carry out tasks such as booking an appointment or transferring the call to a person. It also builds chat agents for text and SMS.
Each agent runs on four layers. Telephony brings the call in by phone number, SIP trunk (a standard connection between phone systems over the internet) or web. Speech to text turns the caller's audio into text as the caller talks. A large language model (LLM, a neural network that generates text) chooses the reply from the agent's prompt and tools. Text to speech sends the reply back as audio. Retell runs all four layers and handles turn-taking and interruptions. An agent is defined by one prompt or by a conversation flow drawn on a visual canvas, then tested with simulated calls and monitored with call analytics and transcripts.
Retell is a hosted service with an API (an interface other programs call over the internet). The standard plan is pay per use, billed per minute of call, with free credit at sign-up; the Enterprise plan has custom prices.