Overview
Vapi is a voice AI orchestration platform for developers who want to build phone and web voice agents without gluing the audio pipeline together themselves. Rather than shipping its own models, Vapi lets you pick each component - speech-to-text from Deepgram or Gladia, any major LLM from OpenAI to Anthropic to Groq, and text-to-speech from ElevenLabs, PlayHT or OpenAI - then handles the hard real-time parts: turn-taking, interruption handling, latency management and session state. It is telephony-native, connecting to Twilio, Vonage or your own SIP trunk, and recently raised a $50M Series B to push further into enterprise. Pricing is usage-based at $0.05 per minute for the platform itself, with model provider costs passed through at cost (or free if you bring your own keys). In practice a production agent lands around $0.15-$0.40 per minute all-in, which is worth budgeting honestly. Enterprise plans add SOC 2, PCI, SSO, RBAC and data residency, with HIPAA available as a $2,000/month add-on.
Key Features
- Real-time voice pipeline orchestration with sub-second turn-taking and interruption handling
- Bring-your-own STT, LLM and TTS with 20+ supported providers
- Native telephony via Twilio, Vonage, Telnyx and custom SIP
- Usage-based $0.05/min platform fee; model costs at cost with your own keys
- 10 concurrent call lines included, scalable for outbound campaigns
- SOC 2, PCI and GDPR compliance with HIPAA and zero-data-retention add-ons
Pricing
| Plan | Price | For |
|---|---|---|
| Build | $0.05/min + model costs (60 free min) | Developers, pay-as-you-go |
| Scale | Custom annual contract | Volume pricing, SLA, dedicated team |
| Add-ons | HIPAA $2K/mo, Zero Data Retention $1K/mo | Regulated industries |
Comparison
Compared to ElevenLabs, Vapi is a full agent orchestration layer rather than a voice model - in fact you can run ElevenLabs voices inside Vapi. Compared to Hume AI, it prioritizes engineering flexibility over emotional expressiveness. If you want no-code agent building instead, Lindy is the friendlier route.
Hands-on Verdict
Vapi is the voice-AI API I use to add conversational voice to my own apps — low-latency calls with LLM orchestration and telephony built in. It’s more developer-grade than Retell on latency and a stronger API than a no-code tool. For no-code calls I use Retell.
Who it’s for: developers, startups. Tip: wire it to your own LLM and a good system prompt — Vapi handles the voice plumbing, so the conversation quality lives in your model and prompt.