Vapi

Summary

Vapi is a developer platform for voice AI agents that make and receive phone calls, orchestrating a speech-to-text, LLM and text-to-speech pipeline on its own infrastructure. Every component is swappable, and Vapi's proprietary turn-taking, interruption and endpointing models run only on its side.

Best for: Engineering teams that want maximum control to build custom voice agents on their own choice of model, voice and telephony.
Not for: Non-technical teams who want a turnkey, no-code voice agent live in an afternoon.

Captured 2026-09-08 · theagentsindex.com

From: $0.05/minutemetered on usage
Free trial: Partly$5 in free credits on signup with no credit card; there is no standing free plan, minimum credit top-up is $10, and 10 concurrent call lines are included on the Build plan.

Visit Vapi Compare Find alternatives

What it does

Done-for-you service: PartlyYou build and run the agent yourself; the Scale contract adds an embedded Vapi engineer for the go-live, not a done-for-you build.A dedicated engineer embedded in your team to get you live in a week.vapi.ai

How much it does unattended

Runs autonomously: YesAgents take whole calls end to end and escalate to a human only when the configured path says so.Automate inbound support calls with agents that can access your knowledge base and escalate to humans when needed.docs.vapi.ai/quickstart/introduction
Multi-agent: YesSquads route one conversation across several specialised assistants with context-preserving handoffs.Squads orchestrate multiple assistants with context-preserving transfers.docs.vapi.ai/quickstart/introduction
Agent permissions: PartlyTools fire mid-call without human approval; a per-tool rejection plan is the mechanism that withholds an action.A rejection plan lets you prevent a tool from executing when certain conditions are met.docs.vapi.ai/tools/tool-rejection-plan

Which models it runs on

Claude: YesClaude is selectable as the assistant's LLM, on Vapi's integration or your own Anthropic account.Anthropic builds the Claude family of large language models. On Vapi, you select a Claude model as the LLM for a voice agent.docs.vapi.ai/providers/model/anthropic
GPT: YesOpenAI GPT models are selectable as the assistant's LLM.Set model.provider to openai and model.model to a GPT model.docs.vapi.ai/providers/model/openai
Open models: YesOpen-weight models reach the pipeline through Groq, Together AI, Mistral, DeepSeek and OpenRouter, or a self-hosted server.Bring your own API keys for transcription, LLM, or TTS — or plug in your own self-hosted models.vapi.ai/platform
Model choice: YesYou choose transcriber, LLM and voice independently and can swap any of them.Bring your own model, or choose from our managed list of 200+ models and integrations.vapi.ai/platform

Where you use it

In your editor: PartlyNo agent runtime in the editor; `vapi mcp setup` teaches Cursor, Windsurf and VSCode the Vapi API so you can build from the IDE.Your IDE's AI assistant (Cursor, Windsurf, VSCode) gains complete, accurate knowledge of Vapi's APIs and best practices.docs.vapi.ai/cli
On the command line: YesThe Vapi CLI covers assistants, numbers, calls, campaigns and logs from the terminal.The Vapi CLI is the official command-line interface that brings world-class developer experience to your terminal and IDE.docs.vapi.ai/cli
In your pipeline: PartlyEvals run as pipeline quality gates via API and CLI; there is no pull-request surface because there is no code to review.CI/CD integration - Automate quality gates in your deployment pipelinedocs.vapi.ai/observability/evals-quicks…
In a browser: YesA hosted dashboard builds and observes agents, and calls embed in your own web app via the Web SDK and widget.Web integration: Embed voice calls directly in your applicationsdocs.vapi.ai/quickstart/introduction

Whose machine it runs on

Self-hosted: PartlySelf-serve is hosted only; on-premise is offered to large enterprises, and the orchestration layer otherwise never leaves Vapi.Yes, we support on-premise deployments for large Enterprises.docs.vapi.ai/faq
Open source: PartlyThe CLI and SDKs are MIT-licensed on Vapi's GitHub; the platform and its orchestration models are closed and hosted.MIT License - see LICENSE file for details.github.com/VapiAI/cli

What it costs to run

How it meters: YesPer call minute for Vapi hosting, plus model provider cost passed through at cost, plus $10 per concurrent line per month.$0.05 / minvapi.ai/pricing
Free tier: YesSelf-serve signup with a $5 credit grant and no card; it is a grant against usage, not a standing free tier.Yes, every new account gets $5 in free credits to start building without the need for credit card.vapi.ai/pricing
API access: YesREST API covers every capability; server and client SDKs wrap it.Every capability is accessible via REST API. Your team owns the integration end to end.vapi.ai/platform
MCP server: YesA hosted MCP server exposes assistants, calls and numbers to another agent, and assistants can also call external MCP servers.The Vapi MCP Server exposes Vapi APIs as tools via the Model Context Protocol (MCP), so you can manage assistants, phone numbers, and calls from any…docs.vapi.ai/sdk/mcp-server
Bring your own key: YesProvider keys for transcriber, LLM, voice and storage; with your key the model cost drops out of the Vapi bill entirely.At cost ($0 if you bring your own API key)vapi.ai/pricing

Buying it for a team

A company can buy it: YesBilling, credits, add-ons and concurrency all belong to an organization, purchased by a member with the Admin role.Purchase call concurrency from Billing & Add-Ons to increase the number of simultaneous calls available to your Vapi organization.docs.vapi.ai/billing/purchase-call-conc…
Seat model: NoNothing is priced per seat: the plan table meters call minutes and concurrent lines only, and members are invited with RBAC roles.10 included + $10 / line / movapi.ai/pricing
Pooled budget: YesOne organization credit balance and one concurrency pool; every inbound and outbound call draws on the same capacity.All inbound and outbound calls share the organization's capacity.docs.vapi.ai/billing/purchase-call-conc…
Admin controls: YesPer-tool rejection conditions withhold an action, API keys are scoped to specific origins and assistants, and prompt guardrails override other instructions.Every API key belongs to one Vapi organization. Choose the key type based on where your application runs, then limit the key to the origins and…docs.vapi.ai/security-and-privacy/api-k…
Audit log: PartlyThe dashboard keeps org-wide API, call and webhook logs and assistant version history; no member-action audit trail is documented.Monitor all API requests and responsesdocs.vapi.ai/debugging
Single sign-on: PartlySAML 2.0 or OIDC with any IdP, but enterprise-only and without SCIM provisioning; the Build column shows a dash.Vapi works with any IdP that supports SAML 2.0 or OpenID Connect (OIDC), including Google Workspace, Microsoft Entra ID (formerly Azure AD), and Okta.docs.vapi.ai/security-and-privacy/sso

What happens to your code

Opt out of training: PartlyThe privacy policy reserves call content for model training and names no toggle; the $1,000/month ZDR add-on is the only documented way to stop it being kept at all.Call recordings, transcripts and logs may be retained to help train and improve the AI models, in accordance with applicable laws.vapi.ai/privacy
Data residency: PartlyUS is the default region; the EU region is frozen to existing accounts and new EU residency is enterprise-only, so regional control comes from your own buckets and models.Vapi's EU support and self-serve growth are frozen until 2027. Existing EU customers can continue using their accounts, but Vapi is not offering new…docs.vapi.ai/security-and-privacy/data-…
Getting out: YesCredits are bought upfront and spent over any time frame rather than lapsing at month end.You can load credits upfront and use them over any time frame that suits your usage patterns - so you're not penalized for underuse in any given…vapi.ai/pricing
Certifications: PartlySOC 2 Type II and a HIPAA BAA sit in the Scale contract, and the plan table shows SOC2 only in that column; HIPAA mode is a $2,000/month add-on on either plan.HIPAA BAA and SOC 2 (Type II) certifieddocs.vapi.ai/enterprise/plans

29 sourced claims on this page. Checked 2026-09-03. How we check.

Alternatives

Where we would send a reader instead, in that listing's own words.

Retell AI
Platform for building, testing and deploying AI voice and chat agents on phone, web and SMS, billed per minute.
Bland
Enterprise voice AI platform for phone agents, running its own self-hosted models.
Synthflow
Enterprise voice AI platform with in-house telephony for inbound and outbound call automation.
Deepgram Voice Agent API
One WebSocket API that runs speech-to-text, LLM orchestration and text-to-speech for real-time voice agents.

Filed under

APIAutonomousClaudeEnterpriseFree tierGPTMulti-agentOpen models

Sources and updates

The pages we read24

Every fact below comes from the vendor. Nothing here is independently corroborated yet.

Coverage elsewhere2

Updates4

2026-09-03Capabilities · Commercial terms · Free tier
2026-08-31Capabilities · Commercial terms · Free tier · Price
2026-08-22Alternatives · Crunchbase link · Documentation link · LinkedIn link · Repository · Summary
2026-08-19Summary