Vapi

Building a voice agent capable of making and receiving real phone calls requires managing speech recognition, a language model, text-to-speech, and telephony infrastructure all at once. Vapi handles this entire stack so developers can focus on business logic.

In practice, each Vapi assistant combines three components: a speech-to-text transcriber, a large language model that generates responses, and a text-to-speech engine. All three are fully configurable across dozens of providers including OpenAI, Anthropic, Google, Deepgram, and ElevenLabs. Inbound and outbound calls go through phone numbers managed within the platform, and the agent can trigger external actions via tools and webhooks.

Vapi is aimed at developers who want to automate phone workflows: customer support, lead qualification, appointment scheduling, medical triage, or order management. The platform offers two building modes: Assistants (a single system prompt with tools) for most use cases, and Squads (multiple specialized assistants with context-preserving transfers) for complex scenarios.

However, Vapi is not suited for non-technical users: access is exclusively through an API, and advanced configuration requires familiarity with concepts such as webhooks, function calling, and third-party provider integrations.