Vapi lets developers build, test and run voice AI agents for phone and web calls, picking their own speech-to-text, LLM and voice providers and paying a per-minute hosting fee on top of those provider costs.
Used this tool? Rate it
Used this tool? Rate it
Vapi is an AI voice agent platform for developers: it calls itself the developer platform for building voice AI agents and takes over the call infrastructure, while the builder decides which speech-to-text, language model and text-to-speech providers sit behind each agent.
The company, Vapi, Inc., grew out of an AI therapist prototype that co-founder Jordan Dearsley built in 2023 before he and University of Waterloo classmate Nikhil Gupta turned the voice infrastructure underneath it into a product. In May 2026, technology press reported a $50 million Series B led by Peak XV Partners at a valuation of around $500 million, citing a person familiar with the matter.
Scale figures come from the company itself. As captured on October 5, 2026, the homepage cites 1 billion calls supported, more than 2.5 million agents launched and more than 1 million developers. In the same press coverage, the chief executive said Vapi processes between 1 million and 5 million calls a day, mostly from enterprise customers.
Vapi says most of the cost of running a voice agent comes from the underlying transcription, LLM and voice providers, with its own per-minute hosting fee added on top.
Vapi builds on two primitives. An assistant is a single agent with one system prompt, tools and structured outputs, while Squads break complex workflows into multiple specialized assistants that hand off to each other during a conversation. The older Workflows builder was retired by August 18, 2026, and existing workflows stopped running from August 19, 2026 onward, so multi-step designs now use Assistants or Squads.
The fastest way to build an AI phone agent on Vapi is the Dashboard path from the phone quickstart; the same steps can also be scripted with the Server SDKs, cURL or the Vapi CLI.
The deployment figures are reported by the company or its customers, not measured independently.
Vapi is a weaker fit for buyers who need an on-premise deployment, because Vapi does not support on-premise deployments.
Each organization belongs to a single region, and its assistants, tools, phone numbers and SIP credentials live in that region.
Vapi pricing has two layers: Vapi's own hosting fee, and third-party costs for transcription, model, voice and transport.
| Package | Price | Concurrent calls | Organizations | Vapi phone numbers | Raw data retention | Support and SLA |
|---|---|---|---|---|---|---|
| No Success Package (Usage only) | $0 to start | 4 | 2 | 1 | 14 days | Discord community, email without SLA |
| Core | $29/month | 10 | 2 | 5 | 30 days | Email, 2-business-day response SLA |
| Pro | 10% of Vapi hosting fees, $999/month minimum | 30 | 10 | 10 | 180 days | Email and Slack, 1-business-day response, 99% uptime SLA |
| Premier | Custom, via sales | Custom | Custom | Custom | Custom | Named team, 1-hour P0 response, 99.9% uptime SLA |
Usage only starts with $5 in free credits and full platform access. Core is the entry package that bundles zero data retention (ZDR) with a response SLA. Pro's fee is a percentage of hosting fees only, so model costs do not raise it, and it adds role-based access control and a dedicated Slack channel. Premier is negotiated with sales and is the only tier that lists SSO, custom RBAC, a named support team and dedicated SIP. These Success Packages apply to new customers; accounts on legacy pricing keep their existing billing structure and terms.
Press coverage places Vapi among a wave of AI voice startups that includes Sierra, Decagon, PolyAI, Bland, Retell and ElevenLabs. Vapi's chief executive says the company differentiates by focusing less on pre-packaged applications and more on the infrastructure and orchestration layer, aimed at enterprises that want control over reliability, compliance and model behavior.
The figures below come from each vendor's own pricing page as captured on October 5, 2026; they cover different bundles, so per-minute numbers are not directly equivalent.
Vapi's latency figures differ by page. The homepage cites sub-500ms latency at scale. The introduction docs cite sub-600ms response times with natural turn-taking. The FAQ says latency is typically around 800 milliseconds for end-to-end voice processing.
On a third-party product launch platform, Vapi held a 4.9 rating based on 25 reviews as captured on October 5, 2026. The platform's AI-generated review summary says production voice work still feels fragile, with users asking for better debugging, observability, interruption handling, latency control and reliability when calls or providers fail. Use of the service is restricted to people at least 18 years old or over the age of majority where they live.
Two pages describe ZDR access differently. The ZDR documentation says it is enabled from the Dashboard and does not require a separate purchase. The pricing page, by contrast, lists ZDR as included with the addition of any success package.
Training language also differs: the Terms of Service were last updated on September 14, 2026, and the Privacy Policy on September 9, 2023. The Terms say customer content is used to improve features such as speech recognition, end-of-turn detection, interruption detection and latency, in a way that does not identify the business or the caller. They also say Vapi does not share customer content with third-party services for their own independent model training. The Privacy Policy says call recordings, transcripts and logs may be retained to help train and improve the AI models, with retention depending on the customer's configuration and deletion on request. For optional or customer-managed model providers, Vapi makes no representation about their training or retention practices and leaves that evaluation to the customer.
Vapi's FAQ states that the company is HIPAA compliant, SOC 2 Type II certified and GDPR compliant; these are the company's own statements rather than independently verified findings.
Vapi starts with $5 in free credits, one free US phone number and no package fee, but every call minute draws hosting and provider costs from prepaid credits.
The documents differ. The current Terms limit use of customer content to non-identifying service improvements and rule out sharing for third-party independent training, while the older Privacy Policy says recordings, transcripts and logs may be retained to help train and improve the AI models.
No. Vapi does not support on-premise deployments; the closest options are the EU region and storing recordings and logs in your own cloud bucket.
