What Vapi does
Vapi is a developer platform for building inbound and outbound voice agents with configurable transcribers, language models, voices, phone providers, tools, and webhooks.
Vapi provides orchestration rather than one fixed voice model. A team can choose phone numbers or SIP, transcribers, LLMs, voices, tools, webhooks, and fallback behavior, then create assistants through a dashboard, API, SDK, or CLI. That flexibility suits products that need to trade accuracy, voice quality, latency, geography, and cost deliberately.
Build pricing currently lists Vapi hosting at $0.05 per call minute and $0.005 per SMS or chat message, excluding model and transport costs passed through at cost; ten concurrent call lines are included and additional lines are $10 monthly. Scale uses an annual platform fee and committed volume. HIPAA and zero-data-retention options are separately listed add-ons, so total cost must include telephony, transcription, LLM, speech, tools, and retention requirements.
Phone automation carries legal and operational risk. Defaults can include recordings, logs, and transcripts; Build retention is listed as 14 days for calls and 30 days for chat. Configure notice and consent by jurisdiction, authentication on webhooks, least-privilege credentials, redaction, opt-outs, do-not-call controls, rate limits, and human transfer. Enterprise SSO and RBAC exist, but SCIM is not currently supported.
How Vapi works
A caller's audio travels through telephony transport to a configured speech-to-text model. Vapi assembles the transcript, assistant prompt, conversation state, and tool results for the selected language model, then sends its response to text-to-speech and back to the caller. Webhooks and tools can read or update business systems. Developers must validate provider behavior, secure endpoints, obtain call consent, and confirm actions.
How to set up Vapi
Approve the call use case
Document inbound or outbound purpose, regions, consent script, identity checks, prohibited topics, retention, transfer, and emergency behavior.
Choose the component stack
Benchmark telephony, STT, LLM, and TTS combinations on representative accents, noise, latency, interruption, and complete per-minute cost.
Secure tools and webhooks
Use managed credentials, signed or authenticated endpoints, static IP allowlisting where appropriate, least privilege, schemas, and idempotency.
Test failure paths
Simulate silence, crosstalk, voicemail, prompt injection, wrong identity, provider outage, tool timeout, duplicate actions, and human transfer.
Launch with tight controls
Cap concurrency and spend, review recordings lawfully, monitor latency and outcomes, rotate keys, and expand only after audited calls pass.
Vapi FAQs
How much does Vapi cost?
Build lists $0.05 per call minute for Vapi hosting plus telephony and model-provider costs. Add-ons, extra concurrency, and Scale contracts change the total.
Does Vapi provide its own speech and language models?
Its core value is orchestration across selectable providers and custom models; provider costs and capabilities vary, and customers can sometimes bring keys.
How long does Vapi retain data?
The current Build table lists 14-day call history and 30-day chat history. Scale and paid zero-retention configurations differ and must be verified.
Is Vapi HIPAA-ready by default?
No. HIPAA is a paid configuration and requires eligible providers, a BAA and correct customer controls. The pricing page currently lists a $2,000 monthly add-on.
Can a Vapi agent take actions?
Yes, through configured tools and webhooks. Every action needs authentication, validation, idempotency, logging, and human approval where consequences matter.
Listing reviewed 2026-07-24. Product details and pricing can change; verify important terms on the provider's website.
Related Meetings AI tools
Related AI guides
Reviews
Tell the community what you made, what worked, and what you wish you knew before starting.