
Vapi
AI AssistantsUpdated November 20, 2026
Developer platform for building, testing and running voice AI agents that answer and place phone calls or talk to users inside apps.
We may earn a commission when you sign up through our links. It does not affect our reviews.

What is Vapi?
Vapi is a developer platform for voice AI agents that answer and place phone calls or talk to users inside web and mobile apps. Every agent chains three layers, speech-to-text, a language model and text-to-speech, and Vapi runs the orchestration, turn-taking, telephony and scaling between them while you choose the providers: Deepgram, OpenAI, Anthropic, Google, ElevenLabs, Cartesia and many others. The problem it removes is months of real-time audio plumbing, since you configure behavior in a dashboard or through the API instead. The company says it has supported over 1 billion calls, and it raised a $50M Series B led by Peak XV in May 2026. The weak spot is predictability: cost and latency depend on the stack you assemble, and the dashboard assumes technical comfort. Pick it if you have a developer and want control over every layer; skip it if you need a ready-made receptionist.
How to start using Vapi?
Create an account at dashboard.vapi.ai. Every new account starts on usage-only billing with $5 in free credits, so you can place real test calls before you buy anything. If you live in the terminal, install the Vapi CLI with the one-line script from the docs, run vapi login, and you can create assistants, manage numbers and start calls without opening the browser at all.
Open Assistants and use the arrow next to Create Assistant to pick a template: Customer Support, Lead Qualification, Appointment Scheduler or Info Collector. Each template ships with a first message and a system prompt. New assistants use the Balanced preset for the transcriber, model and voice, so you only change components when you have a reason. To rewrite the prompt with AI, select Generate in the System Prompt section, describe the change and accept the proposed edits. You can also describe the whole agent in plain language to Composer, an alpha assistant inside the dashboard that builds and configures it for you.
Next, give the agent a phone line. Under Phone Numbers, select Create Phone Number, choose Free Vapi Number and enter a US area code. Your first free number needs no payment method, but it only takes inbound calls. For outbound calls or non-US numbers, import a number from Twilio, Telnyx or DIDWW, or connect your own SIP trunk. Assign the assistant under Inbound Settings, then add tools such as Transfer Call, API Request or a knowledge base query so the agent can actually do something during the call.
Select Publish to create a version; until then your edits stay in draft and do not affect live calls. Use Talk to run a web call in the browser, then dial the number from your phone. Before real callers arrive, run Simulations, where AI testers hold scripted conversations with your agent, and check the call logs and the Call Cost tab for each test. Traffic splitting later lets you send a slice of live calls to a new version and roll back if numbers drop.
Who is it for?
Developers building voice products
You want voice inside your app or a phone line for your SaaS without building streaming audio, interruption handling and telephony yourself. Vapi gives you Web and server SDKs, a CLI, webhooks and config-as-code, so the voice layer behaves like any other API in your stack.
Enterprise customer support teams
Inbound call volume grows faster than headcount, and long hold times hurt satisfaction scores. Vapi powers 100% of Amazon Ring’s inbound support line, according to TechCrunch, and offers SLAs, SSO and named support on its top package.
AI automation agencies: You build receptionists and booking agents for local businesses and need separate workspaces per client. Organizations, GoHighLevel and n8n integrations, plus white-label dashboards from partners listed in the docs, let you run many client agents from one account.
Healthcare and regulated businesses
Patient scheduling and intake calls involve protected data and audits. Vapi documents SOC 2 Type II, PCI and GDPR practices, and offers HIPAA-eligible handling with a BAA as a paid add-on, so routine calls can move to an agent without breaking compliance.
Outbound sales and operations teams
Reminder calls, surveys and lead qualification eat hours of rep time. Campaigns take a CSV of contacts with variables, respect a calling window, detect voicemail and report the result for each contact.
Teams comparing AI models
You want to try a new language model or voice without a rebuild. Swap providers per assistant, bring your own keys, then split live traffic between versions and compare cost, latency and outcomes on the Traffic tab.
Vapi pricing
Pricing has two layers. Usage is billed per minute: Vapi charges a $0.05/min hosting fee, and speech-to-text, the language model, text-to-speech and telephony are passed through at each provider’s listed price with no markup. On top of usage you can add an optional Success Package for support, SLAs, more concurrency and compliance. All prices below are from the official pricing page in USD.
Pre-paid credits at $0.05/min hosting plus model costs. Includes $5 in free credits, 2 organizations, 1 Vapi phone number, 4 concurrent calls, 14 days of raw data retention, the full model catalog, Discord community help and email support without an SLA.
Raises limits to 5 Vapi phone numbers and 10 concurrent calls, keeps raw call data for 30 days, includes Zero Data Retention, and adds email support with a two-business-day response SLA.
Includes 10 organizations, 10 phone numbers, 30 concurrent calls, 180 days of retention, RBAC and Zero Data Retention, a dedicated Slack channel, a one-business-day response SLA and a 99% uptime SLA.
Custom, through sales only. Adds a named support team with an Agent Strategist and a Forward Deployed Engineer Advisor, a one-hour response SLA for P0 issues, a 99.9% uptime SLA, custom RBAC, SSO and dedicated SIP. The docs say the EU region is available only on this package.
Add-ons work on any package: HIPAA-eligible data handling with a DPA and BAA costs $2,000/month, extra concurrent call lines cost $10/line/month, and extra organizations cost $20/month each. Transport is separate. Vapi telephony, SIP, WebSockets and Daily WebRTC are free, while Twilio costs $0.008/min inbound and $0.014/min outbound, Vonage $0.00814/min and Telnyx $0.0055/min. The official calculator estimates 1,000 minutes with Deepgram, OpenAI and ElevenLabs at $82 to $129 a month, roughly 8 to 13 cents per minute before any package.
Vapi key features
Modular voice pipeline: Each assistant picks its own transcriber, language model and voice from a catalog Vapi describes as 200+ models and integrations. You can add your own provider keys, point to a custom OpenAI-compatible LLM, or plug in a custom TTS or transcriber. Fallback plans switch to a backup voice, transcriber or model when a provider fails, so one vendor outage does not drop every call.
Assistants and Squads: A single assistant handles most jobs with one prompt, tools and structured outputs
Squads chain several specialized assistants that hand the caller off with context, which Vapi recommends for triage, order management and multilingual routing. The older visual Workflows builder was retired on August 18, 2026, so new builds go to Assistants or Squads.
Tools and integrations: Built-in tools transfer calls (blind or warm), hang up, send SMS, press DTMF keys and call any HTTPS API mid-conversation. Function tools hit your own server, MCP tools connect Zapier, Composio and Make, and the docs cover n8n, GoHighLevel and Google Calendar. Knowledge bases with a Query tool ground answers in your files instead of the model’s memory.
Telephony and channels: Free US numbers for inbound calls, imported numbers from Twilio, Telnyx and DIDWW, Vonage as a priced transport option, and SIP trunks for your existing PBX or contact center. Outbound Campaigns run batch calls from a CSV with voicemail detection. The same agents also work over WebRTC, WebSockets, a website voice widget and an OpenAI-compatible chat API.
Testing and safe releases: Simulations run AI testers against your assistant with mocked tools, and Evals check responses and tool calls with pass or fail results. Versioning lets you publish, compare and restore configurations, and traffic splitting sends a percentage of live calls to a new version. Bad changes get caught before most callers hear them.
Observability: Call analysis summarizes and scores every finished call, structured outputs extract fields such as booking details into a schema, and scorecards grade calls on your own metrics. Boards build custom dashboards, while monitors open issues and alert your team by email, Slack or webhook when quality drops below a threshold.
Developer tooling: Server SDKs for TypeScript and Python, a Web SDK, a CLI, and a GitOps repository that keeps assistants, tools, simulations and evals as files in git for pull-request review. Composer builds agents from plain-language requests, and Vapi has announced Agent Loop, a waitlist feature that scores calls against a goal and drafts fixes on its own.
Vapi Pros & Cons
Pros
Vapi gives you control over every layer of a voice agent without making you build the real-time infrastructure underneath. You choose the transcriber, model and voice, pay providers at cost with no markup, and swap any of them later without rewriting the integration. The release tooling is mature for this category: versions, traffic splitting, AI-driven simulations, monitors and call scoring cover the gap between a demo and a production line. Compliance options span SOC 2 Type II, PCI, GDPR and a HIPAA add-on, and the documentation is unusually deep, with SDKs, a CLI and config-as-code. Start from the Balanced preset and change one component at a time, so you know which swap moved latency or cost.
Cons
The bill has many moving parts. The $0.05/min headline covers only hosting, and some Trustpilot reviewers call the provider and telephony charges hidden costs once real per-minute totals arrive. Latency also depends on your stack, and Product Hunt reviewers ask for better debugging when a call goes wrong. Free numbers are US-only and inbound-only, so international teams must import numbers and pay a second carrier. Usage-only accounts get Discord and email help without an SLA, and Trustpilot complaints describe slow answers on live incidents. The dashboard also expects technical comfort. Set auto-reload limits on credits and price a full month of calls in the calculator before you launch.
Use Vapi if you want…
- Build phone agents that answer support calls around the clock.
- Swap speech, language and voice models without rewriting code.
- Qualify leads with outbound calling campaigns from a CSV.
- Book appointments by connecting agents to your calendar tools.
- Test every prompt change with simulated callers first.
- Add real-time voice conversations to your web app.
- Route complex calls through specialized assistants in Squads.
- Meet HIPAA and SOC 2 requirements for regulated calls.
Vapi alternatives
Vapi Reviews
Builders on Product Hunt call Vapi one of the fastest ways to put a voice AI product in front of users without wiring speech, transcripts and call state by hand. The same reviewers want deeper debugging for failed calls and clearer latency breakdowns in production.
Developers praise the APIs, quick setup and the freedom to mix language models and voice providers, but still see room for better interruption handling and lower latency in long, complex conversations.
One Trustpilot reviewer found the agent editor frustrating, with settings appearing to reset when other fields changed and the voice search hard to use when picking a voice for a specific language.
Several Trustpilot complaints focus on money and support: the listed per-minute price covers only the platform, LLM and telephony fees push the real rate much higher, and help on live production issues can take days. A satisfied customer on the same site counters with fair pricing and responsive support.
UK users point out that Vapi does not sell local UK numbers, so they import one from Twilio and end up paying two providers for every call.
FAQ
Does Vapi have a free plan?
There is no monthly free tier. New accounts get $5 in free credits and one free US phone number for inbound calls, and every minute after that is billed from pre-paid credits.
How much does a minute on Vapi really cost?
You pay Vapi’s $0.05/min hosting fee plus the transcriber, model, voice and telephony providers at their listed rates. The official calculator puts 1,000 minutes with Deepgram, OpenAI and ElevenLabs at $82 to $129 a month.
Do I need to know how to code to use Vapi?
You can build and test a basic agent in the dashboard from a template or with Composer. Connecting your own systems through function tools, webhooks or SDKs usually needs a developer.
Which languages does Vapi support?
Vapi says agents can talk in English, Spanish, Mandarin and 100+ other languages. Automatic language detection works with Deepgram Nova 3 in Multi mode, Google’s multilingual setting and Gladia.
Can Vapi make outbound calls?
Yes, but not from free Vapi numbers, which are inbound only. Import a number from Twilio, Telnyx or DIDWW, or connect a SIP trunk, then place single calls through the API or batch calls with Campaigns.
Is Vapi HIPAA and SOC 2 compliant?
Vapi is SOC 2 Type II attested and documents PCI and GDPR practices. HIPAA-eligible data handling with a BAA is a $2,000/month add-on, and Zero Data Retention comes with any Success Package.
Can I use my own LLM or provider API keys?
Yes. You can add provider keys for transcription, models and voices, connect a fine-tuned or OpenAI-compatible model through a custom LLM URL, or plug in your own TTS and transcriber.
Can I run Vapi on-premise?
No. The docs state that on-premise deployment is not supported. Vapi runs as a managed cloud service, with a dedicated EU region available on the Premier package.
Similar tools
MyMapAI canvas that turns a prompt, PDF or YouTube link into editable mind maps, flowcharts, SWOTs, genograms and slides, with GPT, Claude
FirstupEnterprise employee communications platform with AI search, AI writing help and targeted delivery to app, email, intranet and signage.
ChatPRDAI workspace for product managers that drafts PRDs and specs from your tools, reviews them like a CPO and feeds them to


