Vapi
by Vapi
Voice AI infrastructure for developers who want control, not a packaged agent
Vapi is a developer platform for building production voice agents that place and answer phone calls or run live inside an app. It orchestrates speech-to-text, a large language model and text-to-speech into one low-latency loop, and sells the plumbing — turn-taking, interruption handling, model fallbacks, telephony and observability — rather than a finished contact-centre product, so engineering teams keep control of prompts, providers and business logic.
Vapi is voice AI infrastructure that assembles speech-to-text, a large language model and text-to-speech into a single low-latency conversational loop and exposes it to developers as an API rather than as a finished contact-centre application. Founded in 2023 by Jordan Dearsley and Nikhil Gupta and publicly launched in March 2024, the San Francisco company deliberately sells plumbing: teams choose their own providers — OpenAI, Anthropic, Google, Deepgram, Gladia and ElevenLabs are all first-class — while Vapi handles turn-taking, interruption handling, model fallbacks, latency balancing, enterprise rate limits across providers, telephony and call orchestration. Two primitives structure the work. Assistants are a single system prompt with tools and structured outputs, aimed at support, lead qualification and booking. Squads chain several specialised assistants together with context-preserving transfers for flows too complex for one prompt. Agents can make and receive phone calls, embed in web and mobile apps through client SDKs for Web, React Native, iOS, Flutter and Python, and reach external APIs and databases through tools; server SDKs cover TypeScript, Python, Ruby, C# and Swift. Vapi advertises sub-500ms average latency and 99.9% uptime for enterprise clients, and reports supporting over one billion calls to date at one to five million per day, with 2.5M+ agents launched by more than a million developers. In May 2026 it raised a $50M Series B led by Peak XV at roughly a $500M valuation — M12, Kleiner Perkins and Bessemer participating — taking total funding to $72M. Amazon Ring evaluated more than 40 vendors before routing 100% of its inbound calls through Vapi.
The engineering team that already owns a call flow — support, dispatch, claims or scheduling — and wants to run voice agents against its own systems and its own choice of STT/LLM/TTS vendors rather than adopt a packaged agent product it cannot inspect.
A production phone agent with sub-500ms turn-taking that you assemble and control yourself, without building the real-time audio, telephony and model-fallback infrastructure underneath it.
At a Glance
- Category
- Audio & Voice
- Pricing
- Usage-based, Contact for pricing
- Target Market
- CTOs, Enterprise Developers, VP Engineering, Product Engineering Teams, Contact Centre Operations Leaders
- Deployment
- Cloud-only, API-based
- Founded
- 2023
- Headquarters
- San Francisco, United States
- Team Size
- 51-200
- Customers
- Over 1 million developers on the self-serve platform and roughly 2.5M agents launched; named enterprise customers include Amazon Ring, Intuit, ServiceTitan, New York Life, Kavak, Instawork and GoHealth
Key Features
- ✓Provider-agnostic STT/LLM/TTS pipeline
Swap OpenAI, Anthropic, Google, Deepgram, Gladia or ElevenLabs per component, so a model regression or price change is a config edit rather than a rebuild.
- ✓Assistants and Squads
Assistants hold one system prompt plus tools; Squads chain specialised assistants with context-preserving transfers so long flows do not lose state at handoff.
- ✓Sub-500ms turn-taking
The platform advertises sub-500ms average latency with natural interruption handling, which is the difference between a usable phone agent and one callers hang up on.
- ✓Model fallbacks and latency balancing
Routes around a slow or failing provider mid-call and holds enterprise rate limits across vendors, so one upstream outage does not drop live calls.
- ✓Tools and function calling
Agents call external APIs, CRMs and databases during a conversation, which is what turns a scripted IVR replacement into something that can actually complete a transaction.
- ✓Client and server SDKs plus MCP connector
Web, React Native, iOS, Flutter and Python clients with TypeScript, Python, Ruby, C# and Swift server SDKs, and a published MCP connector for agent-built agents.
- ✓Enterprise controls
SOC 2, HIPAA and PCI compliance with SSO, OAuth and role-based access control, plus an enterprise SLA and a forward-deployed team.
Capabilities
Use Cases
- •Inbound customer support deflection
Amazon Ring evaluated more than 40 voice vendors and now routes 100% of its inbound calls through Vapi, escalating to humans on failure.
- •Outbound sales and lead qualification
Agents dial prospects, qualify against CRM criteria and book meetings with branching logic, replacing the first-touch layer of an SDR team.
- •Appointment scheduling and confirmation
Handles booking requests, checks live availability and confirms or reschedules with conditional routing, removing the phone tag from service businesses.
- •In-app voice interfaces
Client SDKs embed a live voice agent directly in a web or mobile product, so the conversation never has to traverse the phone network.
- •High-volume operational calling
Marketplaces such as Instawork and Kavak run staffing and logistics calls at a scale that would otherwise need proportional headcount growth.
Ideal For
Best For
- ✓Automating high-volume inbound support calls with agents that read a knowledge base and escalate to a human on failure
- ✓Outbound qualification and appointment-setting where branching logic has to hit a CRM or scheduling system mid-call
- ✓Embedding live voice conversation inside an existing web or mobile product via client SDKs rather than over the phone network
- ✓Teams that must keep their own LLM and TTS contracts and bring their own API keys instead of paying a marked-up bundle
- ✓Enterprises needing SOC 2, HIPAA or PCI coverage with role-based access control and SSO on a voice workload
Not Ideal For
- ✗Non-technical buyers who want a configured agent out of the box — Vapi is deliberately infrastructure, and Sierra, Decagon or PolyAI sell the finished application instead
- ✗Teams that need a single predictable per-minute price: the $0.05/min platform fee excludes STT, LLM, TTS and telephony, so real all-in cost is several times the headline
- ✗Regulated workloads requiring on-premises or VPC deployment — Vapi is cloud-only, and HIPAA is a paid add-on rather than a default posture
- ✗Small teams without capacity to absorb slow support: Trustpilot reviewers repeatedly describe multi-day waits on production incidents
Integrations
Deployment
Market & Ratings
Over 1 million developers on the self-serve platform and roughly 2.5M agents launched; named enterprise customers include Amazon Ring, Intuit, ServiceTitan, New York Life, Kavak, Instawork and GoHealth
Market Analysis
Pros
- ✓Genuine provider neutrality — swapping STT, LLM or TTS is configuration, and bringing your own keys removes the model markup
- ✓Enterprise-scale evidence rather than claims: over a billion calls, 1-5M/day, and a competitive win at Amazon Ring against 40+ rivals
- ✓Fast to a working prototype, with SDKs across Web, React Native, iOS, Flutter, Python and five server languages
- ✓Compliance breadth for a young company: SOC 2, HIPAA and PCI with SSO, OAuth and RBAC on the Scale plan
Cons
- ✗All-in cost is far above the advertised $0.05/min once speech, model, voice and telephony are added — opaque multi-part billing is the most repeated complaint in public reviews
- ✗Support responsiveness is the single most common Trustpilot grievance, with reviewers describing waits of days to over a week on production incidents
- ✗Latency is inconsistent in the field: reviewers report turn-taking sometimes at 800-1000ms and occasionally 3-5 seconds, against an advertised sub-500ms average
- ✗Reliability complaints cluster around post-prototype production use — settings failing to save, integrations breaking after platform changes
- ✗No self-service cancellation; multiple reviewers describe difficulty closing accounts or removing payment details
- ✗Cloud-only with no VPC or on-premises option, and HIPAA gated behind a $2,000/month add-on
Pricing
Build
From $0.05/min
- ✓Usage-based call minutes at $0.05 / min
- ✓SMS and chat at $0.005 / msg
- ✓10 concurrent lines included, then $10 / line / mo
- ✓Model provider costs at cost ($0 if you bring your own API key)
- ✓14 days call-history retention, 30 days chat
- ✓Community Discord and email support
Scale
Contact for pricing
- ✓Annual contract: fixed platform fee plus committed volume
- ✓Volume-based per-minute rates and custom concurrency
- ✓SOC 2, HIPAA, PCI, SSO and RBAC
- ✓Custom data retention
- ✓Dedicated account team, named support engineer, custom SLA
The published $0.05/min is a platform fee only — speech-to-text, the LLM, text-to-speech and telephony are all billed separately at cost, so buyers should model the stack, not the headline, and reviewers consistently flag the multi-part bill as the biggest surprise. Concurrency beyond 10 lines is $10/line/month and chat is $0.005/message. Compliance is priced as add-ons: HIPAA at $2,000/month and Zero Data Retention at $1,000/month, on top of the Scale plan's fixed platform fee and committed annual volume. Bringing your own provider API keys removes the model markup entirely.
Security & Compliance
Connect
Sources
This page was written from 7 sources, 5 on domains other than vapi.ai.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Omilia Cloud Platform
Self-learning agentic voice AI for enterprise contact centers: voice agents, voice biometrics and fraud defense on one platform
Phonely
Voice agents running on Alma, a model trained on ten million real phone calls
Smallest.ai
Sub-second voice AI — small, specialised speech models for real-time enterprise voice agents
Ringg AI
Multilingual voice, chat and WhatsApp agents for high-volume enterprise conversations