What is Vapi?
Vapi is a developer platform for building, testing and deploying AI voice agents that hold real conversations on phone lines and inside apps. It handles the call infrastructure, turn-taking and orchestration, while you choose the speech-to-text engine, language model and voice that power each agent. Common uses include customer support lines, lead qualification and appointment booking. The company says more than one million developers use it, that it has supported one billion calls, and that over 2.5 million agents have been launched.
You work with Vapi through an API, SDKs, a command-line tool and a dashboard. A single assistant combines a system prompt, tools and structured outputs, while Squads link several specialised assistants on one call. Testing tools such as evals, simulations and scorecards help you check behaviour before callers hear it. Customers named on the homepage include Amazon Ring, Intuit, ServiceTitan and New York Life. Vapi was live and offering self-serve sign-up in October 2026, with usage-based billing and optional support packages.
Key Features
- API-first voice agents: Vapi lets developers build voice assistants from a system prompt, tools and structured outputs through its API. The company says agents go from idea to production in minutes, with the platform handling call infrastructure.
- Squads for multi-agent calls: Vapi Squads coordinate several specialised assistants on one call with context-preserving transfers. A booking agent can hand a caller to a billing agent without the caller repeating details.
- Choice of model providers: Vapi connects to dozens of speech, language and voice providers, including OpenAI, Anthropic, Google, Deepgram and ElevenLabs. You can swap a provider in configuration rather than rebuilding the agent.
- Phone and web calling: Vapi handles inbound and outbound phone calls and also embeds live voice conversations inside apps. Its documentation lists Twilio, Telnyx and DIDWW as telephony providers, with call routing, transfer and voicemail detection.
- Tool and API calling: Vapi agents can call your own APIs, databases and business systems during a conversation. This lets an agent check an order, book a slot or update a record while the caller is still on the line.
- Evals and simulations: Vapi includes evals, simulations, scorecards and monitoring for testing agents before and after launch. Coval notes that production calls can be turned into test cases for regression checks.
- Low-latency conversation: Vapi advertises average latency under 500 milliseconds and natural turn-taking on calls. Its documentation gives a sub-600 millisecond target, and independent reviews place real-world figures around 500 to 700 milliseconds.
- SDKs and CLI: Vapi provides SDKs for several platforms and a command-line tool that exposes the platform from your terminal. Developers can script deployments and manage assistants alongside their existing code.
- Enterprise controls: Vapi offers SSO, OAuth and role-based access, plus SOC 2, HIPAA and PCI compliance claims for enterprise buyers. HIPAA is a paid add-on at $2,000 a month on the official pricing page.
Pros
- Provider flexibility: Vapi avoids locking you into one speech or model vendor, so you can pick the best mix for cost and quality. Both 11x and Coval list this open, swap-friendly design as its main strength.
- Low cost to start: Vapi gives new accounts $5 of free credit with no monthly fee on the base tier. That is enough to build and test a prototype agent before you commit any budget.
- Large developer base: Vapi says more than one million developers use the platform and that it has supported one billion calls. A big community means many tutorials, examples and shared fixes for common problems.
- Proven at enterprise scale: Vapi lists customers such as Amazon Ring, Intuit, ServiceTitan and New York Life. Ring reports routing all of its inbound call volume through Vapi agents.
- Built-in testing tools: Vapi bundles evals, simulations and monitoring into the platform rather than leaving testing to third-party tools. Teams can catch regressions in prompts or providers before callers notice them.
- Transparent usage pricing: Vapi passes provider costs through at their listed rates and shows each component separately. You can see exactly how transcription, model, voice and transport add to each minute.
Cons
- Real cost is higher than headline: Vapi advertises $0.05 per minute for hosting, but provider fees come on top. Coval puts typical totals at $0.07 to $0.15 per minute, and 11x estimates up to $0.27 with premium models.
- Steep learning curve: Vapi expects JSON configuration, code and webhook work, so it suits developers more than business users. 11x describes it as unsuitable for non-developers despite the visual builder and Composer.
- Support is slow on low tiers: Vapi gives base and Core users community or email support with up to a two-day response target. 11x cites a 2.4 out of 5 Trustpilot score with complaints about slow replies.
- Compliance costs extra: Vapi charges $2,000 a month for HIPAA compliance and gates higher concurrency behind paid add-ons. Healthcare and regulated teams face a much larger bill than the per-minute rate suggests.
- Latency varies with providers: Vapi chains speech-to-text, a language model and text-to-speech, and each stage adds delay. Coval notes 500 to 700 milliseconds on this cascaded path, which can feel slow on some calls.
- Testing tied to the platform: Vapi evals and monitoring only measure agents running on Vapi itself. Coval flags this as lock-in if you later want to compare platforms with the same test suite.
Who is Using Vapi?
- Developers and startups: Build voice products and prototypes on a pay-per-minute API.
- Customer support teams: Answer inbound calls and route complex issues to staff.
- Sales and lead teams: Qualify leads and book appointments with outbound agents.
- Voice AI agencies: Deploy agents for several client businesses from one account.
Pricing
Vapi charges $0.05 per minute for hosting plus pass-through provider costs, with optional support packages from $29 a month (official prices, October 2026).
| Plan | Price | What you get |
|---|---|---|
| Base (no package) | $0 plus usage | $5 free credit, 2 organisations, 1 phone number, 4 concurrent calls, 14-day data retention, community and email support |
| Core | $29/month plus usage | 5 phone numbers, 10 concurrent calls, 30-day retention, email support with a 2-day response target |
| Pro | From $999/month (10% of hosting fees) | 10 organisations, 10 numbers, 30 concurrent calls, 180-day retention, dedicated Slack support, 99% uptime SLA |
| Premier | Custom | Higher limits, named support team, 1-hour P0 response, 99.9% uptime SLA, SSO, dedicated SIP |
| Usage | $0.05/min hosting plus provider rates | Transcription, model, voice and transport billed at provider cost; HIPAA $2,000/month, extra concurrency $10 per line |
Older reviews such as Coval’s still refer to a Build plan, and estimates of the true per-minute cost range from about $0.07 to $0.27 depending on the models and voices chosen. The official pricing page is the reference here. Run a few hundred test minutes with your chosen providers and read the usage breakdown before you set a budget.
What Makes Vapi Unique?
Vapi treats every part of a voice call as a replaceable component, so one agent can pair Deepgram transcription with an Anthropic or OpenAI model and an ElevenLabs voice, then switch any of them later. It adds multi-agent Squads and built-in evals on top of that open design, which gives developers fine control over cost, quality and latency in a way that closed, all-in-one voice products do not.
Compatibilities and Integrations
- AI providers: OpenAI, Anthropic, Google Gemini, Groq, Deepgram, ElevenLabs, PlayHT and other speech and model vendors
- Telephony: Twilio, Telnyx and DIDWW numbers, plus web calls embedded in apps
- Business tools: Salesforce, HubSpot, Zendesk and Slack, plus custom API tools
- Developer access: REST API, SDKs for several platforms, a CLI and webhooks
Vapi Tutorials
A first working voice agent takes about five steps.
- Sign up at vapi.ai and claim the $5 starting credit.
- Create an assistant with a system prompt and pick its transcriber, model and voice.
- Add tools that call your APIs, such as a booking or lookup endpoint.
- Buy or import a phone number, or embed the agent as a web call.
- Run evals and test calls, then review logs and costs before going live.
How We Rated It
| Criterion | Score | Why |
|---|---|---|
| Accuracy and Reliability | 4.1 | Strong at scale, some bugs |
| Ease of Use | 3.2 | Developer skills required |
| Functionality and Features | 4.7 | Squads, tools, evals, telephony |
| Performance and Speed | 4.0 | Sub-second, provider dependent |
| Customization and Flexibility | 4.6 | Swap any provider or model |
| Data Privacy and Security | 4.2 | SOC 2, HIPAA, PCI claims |
| Support and Resources | 3.3 | Slow help on low tiers |
| Cost-Efficiency | 3.6 | Hidden provider costs add up |
| Integration Capabilities | 4.6 | Many models, CRMs, telephony |
| Overall Score | 4.0 | Average of the nine criteria |
Summary
Vapi suits developers and technical teams that want full control over the models, voices and telephony behind their voice agents. Its main limitation is complexity, both in setup and in pricing, since provider fees can multiply the $0.05 headline rate. Support on the lower tiers is also slow by several accounts. Start with the free credit, log the true cost per minute of your chosen stack, and only then decide whether a support package is worth paying for.
Reviews
Write a Review