What Is Vapi AI? Features, Pricing and Honest Review

Vapi AI is a developer platform for building voice agents, which are programs that answer phone calls and make outbound calls on their own. The company does not train its own speech models. It connects speech recognition, a language model and a synthetic voice into one live call, and your team chooses every part of that chain.

This review explains how Vapi AI works, what the main features do, and what a call really costs per minute once every provider is added up. The short verdict: Vapi AI is a strong choice when you have developers and want control over each layer, and a weak choice when you need a finished product that runs without code and comes with one predictable bill. All figures were checked against the official pricing page in October 2026.

Vapi AI at a Glance

Here is the whole review in one table for readers who only want the answer. The rest of the article explains each row.

If you only care about money, jump to the pricing section, because that is where most reviews and most buyers disagree.

What Is Vapi AI?

Vapi AI is a developer platform for building voice agents that handle phone and web calls. The easiest way to picture it is as a layer that sits between the caller and several other services. Vapi runs the whole conversation: it listens, sends the text to a model, plays the answer back as speech, and keeps the timing tight enough that the exchange sounds like a normal call. The speech recognition, the model and the voice do not come from Vapi. They come from providers you pick.

That is why reviews call it an orchestration layer. Vapi does not sell a ready made receptionist. It sells the parts a developer needs to build one. Every call passes through three technologies: speech-to-text (STT), a large language model (LLM) and text-to-speech (TTS). If these terms are new, the AI terms glossary explains them in plain language.

Who Is Behind Vapi AI?

The founders are Jordan Dearsley and Nikhil Gupta. They started Superpowered in 2020, a meeting notes app that went through Y Combinator, and Vapi was announced in 2023. The company is based in San Francisco. According to public reports, Bessemer led a $20 million Series A in December 2024 and Peak XV led a $50 million Series B in May 2026, which adds up to about $72 million raised.

TechCrunch reported that Amazon Ring chose Vapi after comparing more than 40 vendors, and the company valuation is reported at around $500 million. These facts show a serious business. They say nothing about whether the product is easy enough for your case.

How Does Vapi AI Work?

Every call follows the same three stages, and Vapi handles the speed of the handoffs between them. This is the path of a single conversation.

Step 1: Speech to Text

Audio from the call streams to a transcription service while the person is still talking. Supported providers include Deepgram and AssemblyAI, with settings for language and for deciding when the caller has finished a sentence. That decision matters more than it looks. If the agent waits too long, the call goes quiet. If it reacts too early, it talks over the caller.

Step 2: The Language Model Decides What to Say

The transcript goes to a language model such as one from OpenAI or Anthropic, or to your own server. The model reads the conversation, follows the prompt you wrote and calls outside tools when needed, for example to check a calendar slot or a record in your database. Then it decides whether to keep talking or end the call.

Step 3: Text to Speech

The answer goes to a voice provider such as ElevenLabs, Azure or Cartesia, which turns it into audio and sends it to the caller. If the person interrupts mid sentence, Vapi stops the outgoing audio and goes back to listening.

Phone Numbers, Twilio and Telnyx

Phone calls need a number. Vapi provides one free US number for inbound calls, and you can import more through SIP or through carriers such as Twilio, Telnyx and Vonage. Web calls from a browser work without a number through the SDK.

What Is Vapi AI Used For?

Most teams use Vapi for calls that follow a pattern and happen in large numbers. The common cases are below.

• Customer support: the agent answers routine questions and hands complex cases to a person.
• AI receptionist: answers the phone for clinics, law offices, home service companies and restaurants.
• Appointment booking: checks a calendar and confirms a time during the call.
• Lead qualification: calls new leads, asks a few questions and passes the result to a CRM.
• Outbound reminders: payment, delivery and appointment reminders at scale.
• Voice inside an app: a talking assistant in a mobile or web product.

Vapi AI by Industry

• Healthcare: scheduling and reminders, but HIPAA costs $2,000 per month extra and every provider needs its own agreement.
• Insurance: intake calls and claim status checks, where a wrong answer has a real cost.
• Real estate: first contact with new leads and showing requests.
• Home services: answering after hours and booking jobs.
• Restaurants and retail: orders, opening hours and reservations.

These cases are a good fit for AI agents in business, but a voice agent is one of the harder kinds to run, because every delay is audible to the customer.

Key Vapi AI Features

Assistants and Squads

An Assistant is a single agent with one prompt, one model, one voice and its own tools. A Squad links several Assistants inside the same call, so a caller can move from a front desk agent to a booking agent and then to billing without repeating anything. Context moves with the call. Squads are the closest thing Vapi has to a multi step flow now that the visual builder is gone.

Speed and Natural Conversation

Vapi advertises response time under 500 milliseconds, and the technical documentation mentions under 600. Independent reviews measure 500 to 800 ms in normal conditions, and some Trustpilot users report pauses of several seconds in production. Speed depends on the providers you choose, the length of the prompt and how many tools are called. The platform also handles interruptions, endpointing (detecting the end of a sentence) and backchanneling, the small sounds such as “mm hm” that make a call feel natural.

Tools, Webhooks and Custom Models

Agents can call functions during a conversation and send events to your server through webhooks. You can also connect your own language model through a custom endpoint, and MCP support lets agents reach external tools. This is the main reason developers prefer Vapi over closed platforms: nothing is locked to a single vendor.

Knowledge Base and Call Analysis

You can upload documents such as a menu, a price list or internal Q and A so the agent answers about your business. After a call, Vapi can write a summary, score the outcome and pull out structured data that you send to a CRM or database.

Evals and Simulations

Evals check individual decisions and tool calls. Simulations run full conversations against an agent or a Squad before real customers reach it. For anything that touches payments or appointments, this saves a lot of embarrassing mistakes.

What Changed in 2026: Workflows Retired

According to the Vapi migration notes, the visual builder called Workflows was retired on August 18, 2026, and existing flows stopped working the next day. New projects use Assistants and Squads. Many older articles and video tutorials describe a screen that no longer exists, so always check the publication date when you learn from old material.

How Fast Is Vapi AI? Latency and Voice Quality

Speed decides whether a voice agent feels like a conversation or like a bad phone line, so every review tests it. The numbers vary a lot by source.

The gap is easy to explain. Vapi only controls its own layer. The rest of the delay comes from the transcriber, the model and the voice you choose, plus the length of your prompt and the number of tool calls per turn. A fast model with a short prompt can feel instant, while a large model with three database lookups per answer will not. Voice quality works the same way: the sound comes from the voice provider, not from Vapi. Test your exact stack on real calls before you trust any published number.

Bring Your Own Keys and Developer Tools

Bring Your Own Keys

With your own API keys for Deepgram, OpenAI, ElevenLabs and other providers, you pay those vendors directly and Vapi charges only the $0.05 platform fee. This is the cost transparency that developers praise. It also means several invoices and several dashboards to watch.

API, SDKs and MCP

Everything in the dashboard can be done through the API, and SDKs cover web, iOS, Flutter, React Native and Python. MCP support lets an agent reach outside tools through a standard connection. For an engineering team this is the strongest part of the product, and it is the reason Vapi gets called developer friendly in almost every review.

Languages, Voices, Integrations and Support

These topics rarely appear in a feature list, but they often decide whether an agent survives in production.

Languages and Voices

Vapi lists support for more than 100 languages, but that number belongs to the providers. The language depends on the transcriber and the sound depends on the voice service, so the same language can work well in one setup and badly in another. Voices come from services like ElevenLabs, Azure and Cartesia. For less common languages, record a few test calls with different combinations before you commit.

Integrations

Vapi connects to other systems through webhooks, function calls and servers you host. Many teams pair it with Make, n8n and GoHighLevel automations, and send call data to a CRM, a calendar or a database. There is no large library of ready made connectors like on no code platforms, so someone has to set up and maintain each connection.

Scaling and Concurrent Calls

The entry plan includes 4 concurrent calls, and each extra line costs $10 per month. The Pro plan raises the limit to 30 concurrent calls. For an outbound campaign or a seasonal spike, this is the first limit teams hit, so calculate the number of lines before launch, not after.

Customer Support

The entry plan uses Discord and email with no guaranteed response time, while Slack support and 99% uptime start on Pro. This matches the most common complaint from users: while a project is small, an answer can take days. The documentation and GitHub examples cover the basics well, but harder problems are often solved by searching the community.

Vapi AI Pricing: What a Call Really Costs

Pricing is where reviews disagree the most. Vapi charges $0.05 per minute of call time for hosting and orchestration.

That is not the price of a call. Transcription, the language model, the voice and the phone carrier are billed on top of it, either directly by each provider if you bring your own API keys, or through Vapi at cost. The pass through is honest, but the bill comes from many places. Always confirm current numbers on the official pricing page.

How Much Does Vapi AI Cost Per Minute?

The calculator on the Vapi site, for 1,000 minutes per month with Deepgram, OpenAI and ElevenLabs, gives $0.082 to $0.129 per minute, or $82 to $129 per month. The low end means cheaper models and voices, the high end a pricier model. Independent reviews and competitors quote roughly $0.12 to $0.33 per minute in real use, because a long prompt, an expensive voice and frequent tool calls push the bill up fast. Carrier fees come on top: Telnyx $0.0055, Twilio inbound $0.008 and Twilio outbound $0.014 per minute. For a wider view of what similar services charge, see how much AI tools cost.

Vapi AI Plans: Usage Only, Core, Pro and Premier

Every account pays by usage. Success plans sit on top and add higher limits, longer data retention, support and uptime guarantees.

Add-Ons: HIPAA, Data Retention and Extra Lines

• HIPAA: $2,000 per month per organization, includes a standard data processing agreement and a BAA.
• Extra concurrent line: $10 per month per line.
• Extra organization: $20 per month.
• Vapi phone number: $2 per month.

Older articles still list $10 in credits and $1,000 for HIPAA. That is outdated, so check the date of any review you read.

Is Vapi AI Free?

No. A new account receives $5 in credits, one number and access to all models, with no card required. After the credits are spent, every call is billed. How many minutes the $5 covers depends on the models you pick, since cheaper ones use less. It is enough for several test calls and for measuring what your own setup costs.

Is Vapi AI Expensive?

It depends on the setup. The table below estimates three volumes, once at the calculator range and once for a pricier setup of about $0.25 per minute. The numbers exclude outside telephony and developer time.

Healthcare adds the $2,000 HIPAA fee. At 5,000 minutes on the pricier setup, the total is about $3,250 per month, so the add-on costs more than all the calls together. The biggest hidden cost is developer time, because someone has to maintain integrations, retries and monitoring after launch.

How Hard Is Vapi AI to Learn?

It depends on who sets it up. For a developer the learning curve is gentle: the documentation is clear, SDKs exist for several platforms and a first agent is ready in minutes. For everyone else the picture changes, because every step beyond a basic conversation needs code.

How Long Does Vapi AI Setup Take?

A basic inbound agent takes minutes. A working agent that books appointments and writes to a CRM usually takes days of development, and a production grade one with error handling, testing and monitoring takes longer. Plan time for the integrations, not for the conversation.

Most of the time goes into connecting outside systems, handling errors and testing, not into the conversation itself. That is why some user reports describe dozens of hours spent on a single agent.

Vapi AI Pros and Cons

Vapi AI Reviews: What Users Say

Ratings on Product Hunt, G2 and Trustpilot

Ratings differ more than for most tools: 4.9 out of 5 on Product Hunt (25 reviews), 4.2 out of 5 on G2 (3 reviews) and about 2.4 out of 5 on Trustpilot (15 reviews, mostly one star). That is not random. Two different groups are writing. Product Hunt reviewers are developers who just built an agent, while Trustpilot is where people in production go after a problem with billing or support. The samples are small, so the written comments are worth more than the averages.

What Users Like

• A short path from idea to a first working agent.
• Freedom to choose the transcriber, model and voice for each project.
• Documentation, the web SDK and a clear API.
• Easy connection to n8n, CRM and webshop systems through webhooks.

Common Complaints

• Support: waits of several days for production problems.
• Speed in production: pauses of a few seconds between replies, despite the published number.
• Bill above the headline price: the real amount shows up after launch.
• Dashboard instability: changes that do not save and errors after edits.
• Long development time: one Reddit user reports over 100 hours on a single agent, mostly spent on API requests, data mapping and error handling.

Is Vapi AI Legit?

Yes. The platform works, has investors, a large customer in Amazon Ring, a public status page and current documentation. The question does Vapi still exist comes up in search because older articles describe a product that has since changed, such as the retired Workflows builder. The company is active and releasing updates. Legit and reliable are not the same thing, though. The status page lists many outside providers the product depends on, so an outage at one of them can hit your calls too.

Is Vapi AI Secure? HIPAA, SOC 2 and GDPR

Is Vapi AI HIPAA Compliant?

Yes, but at an extra price. HIPAA support costs $2,000 per month per organization and includes a standard data processing agreement and a BAA with Vapi. Each provider in the chain (transcriber, model, voice, telephony) needs its own agreement, because data passes through all of them. SOC 2 Type II is listed for the company, while SSO and RBAC belong to higher plans. Call retention on the entry plan is 14 days, which is often too short for healthcare and finance teams.

GDPR, Data Processing Agreements and Calling Laws

The documentation states GDPR compliance for EU data. One independent review notes practical limits for smaller companies: the data processing agreement is only available at the enterprise level, and there are no EU servers. European companies should check the terms before sending personal data. For outbound calls in the United States, the FCC ruled in 2024 that AI generated voices count as artificial voices under the TCPA, which requires prior consent from the person called. Inbound calls have lighter rules. For the wider picture of data protection when using AI tools, read the guide on AI data privacy for business.

Compliance by Language Model

The table comes from the Vapi documentation on model providers. The list changes, so check the current state before choosing a model.

Who Should Use Vapi AI?

Teams That Fit Vapi AI

• Developers building a voice feature into their own product.
• Teams that want to switch providers when a better model appears.
• Companies with the capacity to manage integrations, retries, testing and monitoring.
• Buyers who accept that the cost per minute depends on the setup they choose.

Who Should Skip It

• Companies without developers.
• Buyers who want one fixed bill they can plan ahead.
• Customers who need a guaranteed support response on the entry plan.
• Anyone looking for a visual builder, since Workflows was retired.

Is Vapi AI Good for Beginners and No Code Users?

You can build a basic agent from the dashboard: write the instructions, pick providers and make a test call. As soon as the agent needs outside data, a condition or logic with several steps, a developer is required, and since Workflows was retired there is no visual way out. Templates are few and mostly shallow. If you want a no code tool, look at how to build a chatbot without coding or choose a ready made calling platform.

Vapi AI Scorecard

The scores below are an estimate based on public sources and user ratings, not the result of a hands on test. They show at a glance where Vapi AI is strong and where it is weaker.

Vapi AI vs Retell AI vs Bland AI

Competitors publish most of these comparisons, and Retell and Bland sell the rival product, so read them with that in mind.

Vapi AI vs Retell AI

Vapi gives more control and a lower platform fee, but the total depends on the stack you assemble. Retell bundles more into one price and is easier for a team without developers. One test reported a response time of 536 ms for Vapi against 714 ms for Retell, but that is a single third party test, not a general result.

Vapi AI vs Bland AI

Bland is aimed at large outbound campaigns and hosts more of the system itself. Vapi suits teams that want to build their own flow. The choice comes down to whether you want to assemble the parts yourself or buy a finished service.

Best Vapi AI Alternatives

For developers who want even more control, LiveKit is an open source framework that demands more of your own work. Vapi is also sold through AWS Marketplace at the same $0.05 per minute. For teams without developers, common alternatives are Synthflow and services that bundle a voice agent with a phone system and CRM. A guide on how to choose AI tools for business covers the criteria, from price to support.

How to Get Started With Vapi AI

1. Create an account and claim the $5 in starter credits.
2. Create an Assistant and name it after the job it will do.
3. Write the prompt: its role, tone, the steps of the call and what it must never say.
4. Pick a transcriber, a language model and a voice, then run a test call from the dashboard.
5. Add a tool or a webhook so the agent can check a calendar or save data to your system.
6. Upload documents to the knowledge base if the agent must answer questions about your business.
7. Set the call analysis so every call ends with a summary and structured data.
8. Connect a phone number, either the free US number or one imported from Twilio or Telnyx.
9. Run simulations and real test calls, read the call logs and fix the prompt.
10. Launch with a small share of traffic, watch the cost per minute and scale from there.

Keep the first agent small. One job, one prompt and one integration is enough to learn what a minute actually costs in your setup.

Final Verdict: Is Vapi AI Worth It?

Vapi AI is worth it for a team that has developers and wants its own voice system. For $0.05 per minute you get a good orchestration layer, solid documentation and the freedom to swap models, and the providers are billed at cost. It is not worth it for a company that just wants the phone answered next week, because the bill is harder to predict and support on the entry plan can be slow.

Before you decide, work out the total cost and the return, which the guide on AI ROI for business explains, and test your own setup with the $5 in credits. Only then do you know what Vapi AI costs in your case.

Vapi AI FAQs

Is Vapi AI Free?

No. A new account gets $5 in credits, and after that every minute of a call is charged. The platform fee is $0.05 per minute, and the providers are billed on top.

Is Vapi AI Legit?

Yes. It is a funded company with real customers, a status page and active documentation. User reviews are mixed, mainly because of support speed and bills that end up higher than the headline price.

Is Vapi AI GDPR Compliant?

The documentation states GDPR compliance for EU data, but a data processing agreement is reported to be available only at the enterprise level and there are no EU servers. European companies should confirm the terms before sending personal data.

Is Vapi AI Open Source?

Not the whole platform. Vapi is a commercial service, but its SDKs for web, iOS, Flutter, React Native and Python are open source on GitHub. A fully free alternative for the whole stack does not exist, but one can be assembled from open source models.

How Much Does Vapi Cost Per Minute?

The platform fee is $0.05 per minute. With speech to text, a model and a voice added, the official calculator shows $0.082 to $0.129 per minute, and real world estimates run from about $0.12 to $0.33.

Do You Need Coding Skills for Vapi?

For a basic agent, no. You can configure one in the dashboard. For tools, integrations and anything beyond a simple conversation, you need a developer.

Does Vapi Work Outside the US?

Yes, but the free Vapi number is American and covers inbound calls only. For outbound calls and numbers in other countries, import them from Twilio, Telnyx or Vonage.

Who Is the Founder of Vapi?

Jordan Dearsley and Nikhil Gupta. They previously built Superpowered, a meeting notes app, and went through Y Combinator.

What Is the Difference Between Assistants and Squads?

An Assistant is one agent with one prompt, model, voice and set of tools that runs the whole conversation. A Squad links several Assistants in the same call, so the caller moves from reception to booking or billing with the context kept.

Is Vapi HIPAA Compliant?

Yes, with an add-on of $2,000 per month per organization, which includes a data processing agreement and a BAA. You also need agreements with every provider in the chain, since data passes through the transcriber, the model, the voice and telephony.

How Do Vapi Credits Work?

Credits are prepaid money from which the cost of each call minute is deducted. A new account gets $5, and after they run out billing goes to your card. Pro and higher plans have a monthly minimum, so credits behave differently than on the usage only plan.

Are AI Phone Calls Legal?

In most cases yes, with conditions. In the United States the FCC decided in 2024 that AI generated voices count as artificial under the TCPA, so outbound calls need prior consent from the person called. In the EU, GDPR applies as well. Rules depend on the country, so check them with a lawyer before a campaign.

Is Vapi a Finished Product or a Toolkit?

A toolkit. Vapi gives you the orchestration, the dashboard and the API, but you choose and connect the providers and write the logic. If you want a finished receptionist, a managed platform is a better match.

Is There a Free Alternative to Vapi?

Not a complete one. Open source projects such as LiveKit let you build a similar voice agent without a platform fee, but you host it, connect the models and maintain it yourself.

What Languages Does Vapi Support?

Vapi lists more than 100 languages. The real coverage depends on the transcriber and voice provider you choose, so test your language with the exact providers before launch.