Vapi
Developer platform for voice AI agents that handle phone calls
Last reviewed 2026-06-18
Vapi is an API-first developer platform for building voice AI agents that make and receive phone calls and run voice conversations in web and mobile apps. It orchestrates the three components of a voice pipeline (speech-to-text, an LLM, and text-to-speech) into one low-latency real-time service, letting developers pick best-in-class providers per layer or bring their own models and keys. It handles the hard real-time infrastructure: turn-taking, telephony/SIP, WebRTC, latency balancing, and model fallbacks. Vapi offers two build primitives: single-prompt Assistants and multi-assistant Squads with context-preserving transfers. Beyond the core pipeline it provides tool/function calling, MCP support, knowledge integration, observability, evals and test suites, A/B experiments, and enterprise features such as SSO, RBAC, and SOC 2/HIPAA/PCI compliance.
What it can do
Orchestrate STT-LLM-TTS voice pipeline
SupervisedCombines speech-to-text, an LLM, and text-to-speech from many provider integrations into one real-time pipeline targeting sub-500ms latency, with bring-your-own-model support.
sourceHandle phone calls end to end
AutonomousA configured voice agent conducts a full live phone conversation (telephony/SIP, turn-taking, voicemail detection, warm transfers) without a human on the call.
sourceCall tools and APIs mid-conversation
AutonomousAgents trigger function calls, MCP servers, and authenticated internal APIs to fetch data and take actions during the call.
sourceTest, eval, and observe agents
SupervisedAutomated test suites, evals, A/B experiments, logging, and analytics; setup and tuning are human-driven.
source
Strengths
- +Deep modularity and control: swap any STT/LLM/TTS provider, bring your own keys and models, fine-tune thousands of config points
- +Low latency at scale with enterprise reliability features (model fallbacks, SOC 2/HIPAA/PCI), validated by large production deployments
- +Transparent usage pricing with at-cost model pass-through and a low platform fee, plus open-source SDKs and a CLI
Limitations
- −Complexity and steep setup; the no-code builder is limited and non-engineering teams face significant manual configuration
- −Costs can stack unpredictably: the per-minute fee is on top of separately billed provider costs and add-ons
- −Heavily developer-oriented; teams wanting fully managed or visual-first tooling may find gaps
Overview
Vapi is an API-first developer platform for voice AI agents that handle phone calls and in-app voice. It handles the hard real-time infrastructure so developers focus on the agent.
What it does
Vapi orchestrates a speech-to-text, LLM, and text-to-speech pipeline into one low-latency service, targeting sub-500ms responses. A configured agent conducts full live phone calls end to end (autonomous for the call flow), can call tools, MCP servers, and APIs mid-conversation, and supports warm transfers to humans. It also provides evals, test suites, A/B experiments, and observability.
Integrations & setup
Telephony via Twilio and SIP; STT via Deepgram and others; TTS via ElevenLabs and others; LLMs from OpenAI, Anthropic, and Google, or bring your own. Open-source SDKs and a CLI are available.
Pricing
Usage-based: a $0.05/min platform fee, with model provider costs passed through at cost (and $0 to Vapi if you bring your own keys). Enterprise uses a fixed platform fee plus committed volume.
Best for / not for
Best for engineering teams that want deep control over a voice agent and its providers. Less suited to non-technical teams wanting a turnkey, visual-first product.
Alternatives
Sierra and Decagon offer managed voice support agents at the application layer rather than infrastructure.
What people are saying
We aggregate real LinkedIn discussion into sentiment for the agents people search most. Vapi isn't tracked yet, want it added? Request tracking.
FAQ
How does Vapi charge for calls?+
A usage-based platform fee starting at $0.05/min; underlying speech-to-text, LLM, and text-to-speech provider costs are passed through at cost, and are $0 to Vapi if you supply your own API keys.
Can I use my own LLM and voice providers?+
Yes. Vapi is model-agnostic. Choose from many built-in integrations (OpenAI, Anthropic, Google, Deepgram, ElevenLabs, and others) or bring your own keys and self-hosted models for any pipeline layer.
Sources
- Vapi introduction (docs) · accessed 2026-06-18
- Vapi pricing (official) · accessed 2026-06-18
- Vapi valuation, funding & news (Sacra) · accessed 2026-06-18
Last reviewed 2026-06-18