Last updated: 2026-09-24
Vapi is a developer voice AI platform, founded in 2023, that Amazon Ring picked over more than 40 competing vendors in 2026 to route 100% of its inbound customer calls. Teams bring their own LLM, speech-to-text, text-to-speech and telephony provider, and the platform has processed over 1 billion calls to date.
About Vapi
Vapi is a developer-first voice AI platform founded in 2023 by Jordan Dearsley and Nikhil Gupta, launched publicly in March 2024. It sells orchestration infrastructure, not a single fixed voice bot: teams choose their own large language model (OpenAI GPT, Anthropic Claude, Google Gemini, or a custom endpoint), their own speech-to-text and text-to-speech providers, and their own telephony carrier, and Vapi coordinates the real-time turn-taking between them during a live call.
The platform's core technical claim is a custom fusion audio-text model built to recognize natural pauses and mid-sentence interruptions, targeting sub-500ms response latency when paired with a fast provider stack (commonly Deepgram plus Cartesia plus a GPT or Claude model). Assistants can call external tools and APIs mid-conversation, and a native Model Context Protocol integration lets an assistant pull in MCP-server tools, including Zapier and Composio actions, without custom integration code for each one. In December 2025 Vapi shipped an Evals suite for testing voice agents with JSON-defined mock conversations, three judge types (exact match, LLM-as-judge, and tool-call verification), and a CLI for CI/CD pipelines.
The clearest evidence of production fit is Amazon Ring, which switched to Vapi after evaluating the field of AI voice vendors and now routes its inbound customer calls through the platform, according to Vapi's May 2026 funding announcement. Other enterprise customers named in that disclosure include Kavak, Instawork, New York Life, UnityAI, Cherry, and Intuit, alongside a large self-serve developer base. That scale attracted a $50M Series B in 2026 led by Peak XV Partners, with Microsoft's M12, Kleiner Perkins, and Bessemer Venture Partners participating, at a $500M valuation and bringing total funding to $72M.
The self-serve Build plan is pay-as-you-go with no monthly contract, though the underlying STT, LLM, TTS, and telephony costs are billed separately by their own providers on top of the base per-minute orchestration fee. The enterprise Scale plan is an annual contract with a fixed platform fee and volume-based per-minute rates, built for teams that need SOC 2, HIPAA, and PCI compliance with a dedicated account team. That flexibility comes with a tradeoff reviewers raise consistently: G2 and Trustpilot users report a steep learning curve for non-developers and intermittent latency spikes once an agent moves into production, a different failure mode than the more managed, single-vendor stacks sold by competitors like Retell AI.
Pricing
005 per SMS or chat message, billed to the second, no monthly commitment or contract. Includes 10 concurrent call lines, then $10/month per additional line. New accounts get $10 in one-time free credits (roughly 150-200 minutes of testing).
31/minute. Scale plan (enterprise): annual contract with a fixed platform fee plus volume-based per-minute rates, quoted per contract, including SOC 2, HIPAA, PCI, SSO and a dedicated account team. HIPAA mode costs an extra $2,000/month and the Zero Data Retention add-on costs an extra $1,000/month on either plan.
| Tier | Monthly price | What it includes |
|---|---|---|
| Build (pay-as-you-go) | Free | |
| Scale (Enterprise) | Custom |
Key Features
- Bring-your-own-stack orchestration: Developers pick their own LLM (GPT, Claude, Gemini, or a custom endpoint), STT provider, TTS voice, and telephony carrier, and Vapi coordinates the real-time call between them instead of locking teams into one vendor's stack.
- Sub-500ms turn detection: A custom fusion audio-text model reads audio and transcript together to catch the exact moment a caller finishes talking or tries to interject, keeping round-trip response latency under 500ms with a fast provider stack.
- Native MCP tool access: Assistants connect to any Model Context Protocol server mid-call, including Zapier and Composio, to use thousands of pre-built tool actions without writing a custom integration for each service.
- Evals test suite with CI/CD hooks: Shipped December 2025: JSON-defined mock-conversation tests with exact-match, LLM-as-judge, and tool-call-verification checks, run from a CLI wired into CI/CD pipelines.
- Vapi CLI scaffolding: Generates production-ready webhook handler code (for example /pages/api/vapi/webhook.ts) and forwards webhooks to a local server for development testing.
- Enterprise compliance add-ons: SOC 2 Type II and PCI are included on the Scale plan; HIPAA (with a signed BAA) and Zero Data Retention are both available as separately priced add-ons for regulated workloads.
Pros
- Amazon Ring dropped its previous voice vendors for Vapi and now sends all of its inbound customer calls through the platform, a rare named enterprise reference in this category.
- More than 1 million developers build on the self-serve platform, and the company reports it has processed over 1 billion calls in total to date.
- No fixed monthly contract on the self-serve Build plan: usage starts at $0.05/minute with 10 concurrent call lines included and a $10 signup credit to test with.
- $72M raised in total, including a 2026 round that brought Peak XV Partners, Microsoft's investment arm, and two other venture firms onto the cap table.
Cons
- Trustpilot rating is 2.4/5 across 15 reviews, with recurring complaints about latency spikes reaching 4-5 seconds and slow support once an agent reaches production scale.
- Native phone numbers are only available in the US and Canada; teams elsewhere must bring their own telephony carrier.
- The dashboard's visual builder is limited for production use and the platform has a steep learning curve, effectively requiring a developer rather than a no-code team member.
- HIPAA mode ($2,000/month) and Zero Data Retention ($1,000/month) are both paid add-ons on top of usage costs, not included by default.
Data Handling
- Training-data policy
- Call recordings, transcripts and logs may be retained to help train and improve Vapi's models unless the customer enables the paid Zero Data Retention add-on. Default retention varies by plan, from about 7 days on pay-as-you-go to 30 days on other plans, and data is deleted on request.
- Compliance
- SOC 2 Type II · HIPAA (BAA available · paid add-on) · PCI · GDPR
Frequently Asked Questions
What does Vapi actually cost?
The self-serve Build plan is pay-as-you-go at $0.05 per minute of call orchestration plus $0.005 per SMS or chat message, with no monthly contract and 10 concurrent lines included before $10/month per extra line. Underlying speech-to-text, LLM, text-to-speech and telephony costs are billed separately, bringing realistic all-in cost to roughly $0.07-$0.31/minute. The enterprise Scale plan adds a fixed annual platform fee with volume-based rates, and HIPAA ($2,000/month) or Zero Data Retention ($1,000/month) cost extra on either plan.
Can you use Vapi without paying?
Vapi has no ongoing free tier: new accounts get $10 in one-time signup credits, which covers roughly 150-200 minutes of testing before usage billing kicks in. There is no free forever plan for production use.
What are Vapi's closest competitors?
Retell AI is the closest direct rival, offering a more managed, lower-configuration phone-agent platform where Vapi trades ease of setup for provider flexibility. ElevenLabs' Conversational AI product competes for teams that want one vendor's voice stack end to end, and Deepgram is a fit for teams that only need the speech-to-text layer rather than a full orchestration platform.
How does Vapi compare to Retell AI in 2026?
Retell AI wins on default latency and time-to-production with a managed stack, typically landing around $0.07/minute all-in with HIPAA support as standard. Vapi wins on flexibility: it lets a team swap the LLM, STT, TTS and telephony provider independently, which suits engineering teams building custom voice infrastructure at scale but adds configuration overhead Retell avoids.
How do you set up Vapi?
Sign up on vapi.ai, then use the dashboard or the Vapi CLI to create an assistant by choosing an LLM, a speech-to-text provider and a text-to-speech voice. From there you connect a phone number or web widget, define any tool functions the assistant can call (including MCP servers), and test the call flow before going live; the CLI can scaffold a working webhook handler for local development.
Top Alternatives
- Retell AI: Pick Retell AI if you want a managed, low-latency phone agent out of the box; pick Vapi if you need to swap the LLM, speech-to-text, text-to-speech or telephony provider yourself.
- ElevenLabs: Pick ElevenLabs if you want an all-in-one Conversational AI stack from a single vendor; pick Vapi if you want to use ElevenLabs as just your voice layer inside a multi-provider build.
- Deepgram: Pick Deepgram alone if you only need speech-to-text; pick Vapi if you want the STT, LLM, TTS and telephony pieces orchestrated together into a working phone agent.