AgentVoice builds AI voice agents on Vapi for clients who like its flexibility but don’t want to staff the engineering it takes. Vapi pricing is easy to underestimate, because the $0.05 a minute on its pricing page, as of September 2026, covers only Vapi’s hosting fee. Below we add up the full cost of running an agent, walk through the work we do on Vapi, and name the Vapi alternatives we’d choose when a project needs something Vapi doesn’t offer.
What Vapi is and who it suits
Vapi calls itself “the developer platform for building voice AI agents,” and its FAQ says it “focuses on developers.” You assemble an agent from parts, picking the speech-to-text, language model and voice providers, writing the prompt, and connecting the tools the agent calls during a conversation. Nearly every layer can be swapped for another provider or your own, and when you add your own provider keys, those providers bill you directly.
The way agents get built changed this summer. Vapi retired its visual Workflows builder on August 18, 2026, and existing workflows stopped running the next day. Agents are now Assistants, each a system prompt with its tools, or Squads of assistants that hand a call to one another. Composer, an AI-assisted builder in the dashboard, is still in alpha.
That setup suits developer-led teams that want control over every provider, need a choice of US or EU hosting, and have someone to maintain prompts, tool endpoints and tests over time, on their own staff or on ours. A business without engineers can still run an agent on Vapi as long as someone owns that upkeep after launch.
Vapi pricing as of September 2026
Vapi bills by the minute. Each minute an agent runs carries Vapi’s $0.05 hosting fee plus the cost of transcription, the language model, the voice and transport, which is the phone or web connection. Vapi passes those four through “at the provider’s listed price with no markup,” or you connect your own keys and pay each provider yourself.
On top of usage, new Vapi accounts can add an optional package, and Vapi says accounts that existed before the packages launched see no change for now. The table lists what each package includes, taken from Vapi’s pricing page in September 2026.
| Package | Price | Concurrent calls | Raw data retention | Support |
|---|---|---|---|---|
| Usage only | $0 to start, $5 free credits | 4 | 14 days | Discord community, email without an SLA |
| Core | $29/month | 10 | 30 days | Email with a 2-business-day response SLA |
| Pro | 10% of the hosting fee, $999/month minimum | 30 | 180 days | Email and a dedicated Slack channel, 1-business-day SLA, 99% uptime SLA |
| Premier | Contact sales | Not listed | Not listed | Named support team, 1-hour P0 response SLA, 99.9% uptime SLA |
Add-ons are billed separately. HIPAA mode costs $2,000 a month per organization and includes Vapi’s standard Data Processing Agreement and Business Associate Agreement. Extra concurrent call lines are $10 each per month and extra organizations are $20 each, and Vapi’s docs note that its Zero Data Retention setting can’t be combined with HIPAA mode.
What a Vapi agent costs to run
Vapi’s own calculator is a fair baseline. At its default of 1,000 minutes a month, with Deepgram transcription, an OpenAI model, an ElevenLabs voice and Vapi telephony, it estimates $50 in hosting, $10 for transcription, $8 to $45 for the model and $15 to $24 for the voice. That totals $82 to $129 a month, or roughly 8 to 13 cents a minute before any package or add-on.
The model is the widest band in that estimate, and it’s where our cost tuning starts. The calculator shows OpenAI models running from under a cent to about 4.5 cents a minute, so we match the model to the decisions each call has to make. Transport can move the number as well. Vapi telephony and SIP connections are listed as free, while Twilio bills $0.008 a minute inbound and $0.014 outbound and Telnyx bills $0.0055.
Fixed monthly costs sit on top of usage, and on a small agent they can outweigh it. If a clinic runs 5,000 minutes a month at 10 cents a minute, usage comes to about $500, and HIPAA mode adds $2,000 to that. More simultaneous calls raise the fixed cost too, through a package or through extra lines at $10 each.
What we build on Vapi
On Vapi we build phone agents that carry a call through to a finished task. We write the assistant’s prompt and call logic, build the function tools that reach your systems, and test the agent on real calls before launch. Depending on the job, that covers:
- Screening spam calls and identifying callers before they reach your team
- Answering questions from your documents and web pages through Vapi's knowledge base tool
- Updating your CRM and tagging leads during or after the call
- Booking appointments and texting a confirmation with Vapi's SMS tool
- Transferring callers to staff, with a warm handoff on Twilio lines
- Triggering follow-up actions and a call report once the call ends
For outbound work we use Vapi’s Campaigns, which hold up to 10,000 contacts each and can be scheduled up to seven days ahead. Campaigns need your own numbers, since Vapi’s free numbers aren’t supported for them, and voicemail detection lets the agent leave a message when nobody picks up. When one prompt gets too long to hold every path, we split the agent into a Squad, for example one assistant for intake and another for scheduling, so each stays short enough to test properly.
Vapi includes testing and simulation with every plan, and we use its evals to rerun known calls before a prompt change goes live. If recordings need to stay in your own storage, Vapi can write them to your AWS S3, Google Cloud or Cloudflare R2 bucket, and we set that up during the build. Vapi usage is billed to your own Vapi account, and we quote the build itself after a scoping call, as we do for custom builds on any platform.
When we'd recommend a Vapi alternative
Your team wants to edit call flows visually
With Workflows retired, Vapi no longer has a visual flow builder, and Composer is still in alpha. Retell AI’s Conversation Flow is a working drag-and-drop editor, and ElevenLabs Agents has a visual Workflows editor, so we’d build on one of those when people outside engineering will adjust call paths. Our Vapi vs Retell comparison covers the other differences between those two.
You need a BAA on a modest budget
Retell AI’s BAA is self-signed with “no additional fee” on pay-as-you-go, against $2,000 a month for Vapi’s HIPAA mode. For a practice with a few thousand minutes of calls a month, that difference is larger than the usage bill.
You want one rate for the whole stack
If a predictable invoice matters more than provider choice, Bland AI’s per-minute rate includes the model, transcription and voice. It’s $0.14 a minute on Bland’s Start plan and $0.12 on Build, which adds a $299 monthly platform fee. Bland runs its own models and voices, so there are no provider choices to make, and we cover the trade-offs on our Bland AI page.
Call data has to stay in your own cloud
Vapi’s FAQ says it “does not support on-premise deployments.” It runs in US and EU regions that Vapi operates, and recordings can go to your own storage bucket, but the platform itself runs on Vapi’s infrastructure. When call data has to sit in infrastructure you own, ElevenLabs offers private deployments in your AWS or Google Cloud account, and Bland offers VPC and on-premise deployment on its Enterprise plan.
The third route is to license the AgentVoice engine and have us deploy it into your own Azure or AWS account or onto your own servers. The source code sits in your GitHub organization, records and recordings stay in your database under your retention policy, and cloud and AI costs are billed at cost with no per-minute platform fee.