Vapi
Voice agent platform for building, testing, and deploying phone and web voice agents: telephony infrastructure, tool calling, real-time monitoring, and self-serve pricing for developers.
Vapi is a developer platform for building voice AI agents that handle phone and web conversations. It runs the real time loop between a phone system and the AI models: it listens to a caller, decides what to do and speaks back, with interruption handling. Agents answer on Vapi, Twilio, Telnyx or SIP numbers, in web calls and over SMS chat, and outbound campaigns dial contact lists on a schedule.
Every assistant is assembled from a transcriber, a language model and a voice, each chosen from supported providers with the customer's own keys, or the customer's own server acting as the LLM, with fallback plans that switch providers on failure. During a call an agent can call the customer's functions and APIs, use MCP tools, transfer the call to a person, send SMS and search the customer's uploaded knowledge base files. Squads split a call across specialized assistants that hand off to each other with configurable context; Squads replaced the older Workflows builder in August 2026.
Evals test assistants and squads with mock conversations and judged checks, and Simulations run full conversations against AI testers. Call, chat, session, webhook and API request logs record what each agent did, and monitors flag quality drops. Everything is reachable through an API, server and web SDKs, a CLI and the Vapi MCP Server.
Vapi offers SSO through SAML or OIDC, Admin, Editor and Viewer roles, a SOC 2 Type II report and public SOC 3, PCI DSS, a HIPAA mode and zero data retention, with organizations hosted in a United States or European Union region. Pricing is a per minute hosting fee with each provider billed at cost, plus optional success packages.
Vendor details
Canonical URL
https://vapi.ai
Category
Voice agent
Company status
independent
Use cases & customers
Target customers
Deployment options
Integrations
Function tools and API Request tools call the customer's systems; MCP tools, built in transfer, SMS, end call and DTMF tools; Google Calendar, Make and n8n integrations. Telephony on Vapi numbers, Twilio, Telnyx, Vonage and SIP trunks; web calls and SMS chat. REST API, server SDKs, Web SDK, CLI and the Vapi MCP Server.
In practice
Your support line is flooded with repetitive inbound calls. With Vapi you build a voice agent that listens, looks up the caller's account through a custom tool, and resolves or routes the call, scaling to high volume without more staff.
You don't want to be locked to one speech or language provider. Vapi assembles each agent from a swappable transcriber, model, and voice, so you can tune latency and cost and switch components without rebuilding.
A single call spans scheduling, billing, and triage, which shouldn't share one prompt. Vapi's Squads let specialized agents collaborate on the call and hand off context as the caller moves between tasks.
Sources & related URLs
Agentic Index coverage score
11.5 / 14 capabilities · 82%
| Integrations & Tool Calling | Full |
|---|---|
|
Documented custom tool support and prebuilt action tools. Function tools call the customer's server logic and API Request tools call any HTTPS API directly, with credentials held by reference; MCP tools connect an assistant to external MCP servers at runtime; built in tools transfer calls, end calls, send SMS and press DTMF keys; prebuilt integrations include Google Calendar event creation, Make and n8n. Deepgram, OpenAI and other providers are model layers, not integrations. SourceVapi, docs.vapi.ai (Function tools, Integrations, Built-in tools)read 2026-09-27 |
|
| Workflow Orchestration | Full |
|
Multi agent routing the buyer configures. Squads split a call across specialized assistants that hand off through handoff tools with declared destinations, resolved at runtime, with configurable context passing, variable extraction and rejection rules; the squad examples route by verification and inquiry type. Vapi's older Workflows builder was retired on 18 August 2026 and migrated to Squads. SourceVapi, docs.vapi.ai (Introduction to Squads, Handoff tool, Workflows migration guide)read 2026-09-27 |
|
| Knowledge Grounding & RAG | Full |
|
A maintained store over the customer's documents. Knowledge bases are the customer's uploaded files (txt, pdf, docx, csv, md and more, through the dashboard or the file API) kept in Vapi and attached to assistants; the query tool searches them by file ID during a call, with Google Gemini models doing the retrieval, and new knowledge is added by uploading a file, not retraining. A custom knowledge base can swap in the customer's own retrieval server. SourceVapi, docs.vapi.ai (Knowledge bases, Query tool)read 2026-09-27 |
|
| Human Oversight & Guardrails | Partial |
|
Configurable constraints and escalation, no approval step. A tool rejection plan attaches conditions to any tool and blocks the call when the recent conversation state matches them; handoff rejection rules block unwanted transfers between assistants; transfer tools hand a caller to a person by blind or warm transfer. Escalation hands the conversation over rather than holding an action for sign off. SourceVapi, docs.vapi.ai (Tool rejection plan, Transfer Call tool)read 2026-09-27 |
|
| Security, Identity & Governance | Full |
|
Access and compliance are both documented. For access, Vapi supports SSO through any SAML 2.0 or OIDC identity provider (Enterprise or Premier) and RBAC with Admin, Editor and Viewer roles assigned per member, and states that SCIM is not supported. On compliance, Vapi's SafeBase trust center announces a completed SOC 2 Type II examination for 1 July 2025 to 31 July 2026 (report under NDA) and a public SOC 3, and lists HIPAA, PCI DSS v4.0.1, GDPR and CCPA; HIPAA mode and zero data retention are organization settings. SourceVapi, docs.vapi.ai (Single Sign-On) and security.vapi.airead 2026-09-27 |
|
| Observability & Auditability | Full |
|
Run level records of what each agent did. Call logs hold the transcript, messages, analysis, costs, latency, recordings and downloadable pipeline logs per call, with chat, session, webhook and API request logs beside them; monitors evaluate call data against thresholds and open issues; raw data retention runs 14 to 180 days by package. SourceVapi, docs.vapi.ai (Call logs, Logs overview, Monitoring quickstart) and vapi.ai/pricingread 2026-09-27 |
|
| Memory & State Persistence | Partial |
|
Session state with a stated lifetime, short of a memory layer. For chat and SMS, a sessionId groups chats under one persisted conversation that expires 24 hours after creation by default (expirationSeconds), is readable in session logs and can be deleted through the API; within a call, squads carry context across handoffs. Nothing persists what the agent learned about a caller beyond the session; this is conversation state inside one session. SourceVapi, docs.vapi.ai (Session management, Session logs)read 2026-09-27 |
|
| Deployment & Data Residency | Full |
|
Named regions offered as an option. Each organization belongs to the United States or the European Union region, with its own dashboard, API and SIP hosts (api.eu.vapi.ai, sip.eu.vapi.ai), every resource kept in region and each region an isolated environment; the service is hosted only, with no self hosted or customer tenant option documented. SourceVapi, docs.vapi.ai (EU region and data residency)read 2026-09-27 |
|
| Prebuilt Agents, Templates & Packs | Partial |
|
Worked examples the customer assembles. The Guides page offers cloneable builds for appointment scheduling, medical triage and scheduling, ecommerce order management, property management routing, lead qualification and multilingual support, each a different job, but each built step by step by the customer rather than adopted as a packaged agent; Model Intelligence presets are settings. SourceVapi, docs.vapi.ai (Guides)read 2026-09-27 |
|
| Triggers & Channel Coverage | Full |
|
Work can begin without a person starting it, and agents answer across many channels. Outbound campaigns carry a schedulePlan with an earliest and latest dispatch time, so Vapi starts dialing a contact list on its own schedule, with pre dial webhooks; assistant hooks run actions on call events. Agents also answer on Vapi, Twilio, Telnyx and SIP numbers, in web calls, and over SMS chat. SourceVapi, docs.vapi.ai (Campaign scheduling and lifecycle, Assistant hooks, SIP introduction, SMS chat)read 2026-09-27 |
|
| Model Flexibility & Routing | Full |
|
The customer chooses every model. Each assistant's transcriber, model and voice can be swapped for any supported provider, with the customer's own provider keys, Azure OpenAI and Bedrock credentials, or the customer's own server acting as the LLM; transcriber and voice fallback plans switch providers automatically on failure. SourceVapi, docs.vapi.ai (Core Models, Provider Keys, Custom LLM, fallback configuration)read 2026-09-27 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
A documented API and SDKs for Vapi's own platform. The API reference covers assistants, squads, calls, chats, campaigns, sessions and phone numbers; server SDKs, a Web SDK and a CLI sit beside it; and the Vapi MCP Server at mcp.vapi.ai, authenticated by bearer token, manages assistants, phone numbers and calls from any MCP client. SourceVapi, docs.vapi.ai (API reference, Vapi MCP Server)read 2026-09-27 |
|
| Testing, Debugging & Optimization | Full |
|
An evaluation harness on the customer's own agents. Evals test assistants and squads with mock conversations, exact, regex or AI judged checks and tool call checks, returning pass or fail; Simulations run full conversations against AI testers with success criteria and tool mocks; the testing guides set a release and regression cadence. SourceVapi, docs.vapi.ai (Evals quickstart, Simulations overview, Run and maintain tests)read 2026-09-27 |
|
| Browser & Computer Use | Not documented |
|
Vapi documents no browser, desktop or computer control. IVR navigation presses DTMF keys to walk a phone menu, which is a telephone keypad rather than a screen; Web SDK client side tools trigger side effects in the customer's own web app, which is the product running in a browser. SourceVapi, docs.vapi.ai (llms.txt, IVR Navigation, Client-side tools)read 2026-09-27 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Recent platform changes
Agent Phone, now in public beta from Vapi, is a remote MCP server that lets assistants such as Claude Code and Cursor place phone calls for a user. After a one time approval in the browser, the agent dials, works through phone menus, waits on hold, talks to the person and reports back, with no telephony setup.
Bears on: Agent capability
View sourceVapi added a serviceTier setting that lets assistants request faster OpenAI processing for supported GPT-5.6 and GPT-5.5 models. Its September 21 update also adds provider request IDs to call logs for language model, transcription, and voice services when customers bring their own keys.
Bears on: Observability / auditability
View sourceVapi's Composer can now generate simulation scenarios, tester personalities and full test suites automatically, and assistants can fall back to another Anthropic model on Bedrock from the same provider.
Bears on: Observability / auditability
View sourcePricing
Usage based · $0.05/min hosting fee + provider costs; success packages from $29/mo
usage
Included quota
Without a package: $5 free credits, 2 organizations, 1 Vapi phone number, 4 concurrent calls, 14 days raw data retention. Core ($29/mo): 5 numbers, 10 concurrent calls, 30 days retention, ZDR. Pro ($999/mo minimum, 10% of hosting fee): 10 organizations, 30 concurrent calls, 180 days retention, RBAC and ZDR, 99% uptime SLA. Premier: custom concurrency and retention, SSO, dedicated SIP, 99.9% uptime SLA.
What is public
The hosting fee, a cost calculator by provider, success package prices (Core $29, Pro $999 minimum), transport rates and add on prices are published; Premier is by sales.
Billing mechanics
Per minute hosting fee plus pass through provider costs for transcriber, model, voice and transport; customers can connect their own provider keys.
Cost watchouts
The hosting fee is one layer of the per minute cost: transcriber, model, voice and transport bill separately at provider cost, so the all in rate depends on the providers chosen. HIPAA is a $2,000 a month add on. Pro carries a $999 monthly minimum.
Variable cost rationale
Five separately metered billing layers scale with call minutes and model choice; identical call volumes yield very different bills depending on the stack.
Additional watchouts
Model the full stack in Vapi's calculator before committing; the hosting fee alone understates the all in rate.
Overage / add-ons
Every minute bills the $0.05 hosting fee plus provider costs for each layer; extra concurrency $10 a line a month and extra organizations $20 a month on any package.
Sales call required
Mixed (some tiers require a call)
Free / trial
$5 free credits to start; usage based thereafter
Lowest paid plan
Pay as you go at $0.05 a minute hosting fee plus provider costs; Core success package $29 a month
Commercial notes
Self serve usage pricing, with optional success packages that add support SLAs, longer retention, RBAC, ZDR and, on Premier, SSO.
Key ambiguities
All in per minute cost depends on the providers chosen for each layer; Premier pricing is not published.
Missing data
Premier pricing is not published.
Related vendors
- aiola — Voice agents for field teams on Salesforce: reps speak, and a squad…
- AviaryAI — Outbound AI voice agents for credit unions, banks and insurers,…
- Avoca — AI front office for home services: agents answer every call, text…
- Broccoli — AI voice agents built for the trades, answering every call for…
- Callers — Callers (formerly Voxia): AI agents for inbound and outbound calls,…
- careCycle — careCycle offers AI employees and a softphone platform for Medicare…
Alternatives to Vapi
The closest documented capability profiles to Vapi among voice agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Smallest AI11.0 / 14A lighter documented profile than VapiVapi vs Smallest AI →
- Gnani.ai11.5 / 14Fuller documented coverage on Prebuilt Agents, Templates & Packs
- Phonely11.5 / 14Fuller documented coverage on Prebuilt Agents, Templates & Packs
- Regal10.5 / 14A lighter documented profile than Vapi
- Synthflow12.5 / 14Fuller documented coverage on Memory & State Persistence and Prebuilt Agents, Templates & PacksVapi vs Synthflow →
- PolyAI11.0 / 14Fuller documented coverage on Memory & State Persistence
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded