Vapi vs ElevenLabs Agents: which voice platform fits?

How do Vapi and ElevenLabs Agents compare?

Both host voice agents, offer a website widget, an API and an MCP server, and take phone calls. They differ in who supplies the models, how knowledge stays current and how they charge. We checked both on their official pages on 8 October 2026.

How do the Vapi and ElevenLabs widgets compare?

Vapi’s widget is a vapi-widget tag loaded from a script. You choose voice mode or chat mode, a light or dark theme, one of four corners, custom labels and an optional consent form. It needs a public API key, which Vapi says to restrict to the origins and assistants that use it.

ElevenAgents’ widget is an elevenlabs-convai tag with an agent ID. It runs voice only by default, can let visitors switch between voice and text, or can start in chat, and you style it in the dashboard or with HTML attributes. Widgets need a public agent with authentication turned off, and ElevenLabs recommends an allowlist of your domains.

Both also have SDKs for building your own interface. Kawl Now’s widget is voice only: a visitor types a topic, allows the microphone and talks, with captions, mute and an end button.

Who supplies the models and voices?

Vapi is assembled from providers. Each assistant pairs a transcriber, a language model and a voice from providers such as Deepgram, OpenAI, Anthropic, Google and ElevenLabs, so a Vapi agent can speak with an ElevenLabs voice. Vapi charges $0.05 a minute for hosting and passes model costs through without markup. Calls in several languages depend on choosing a transcriber that detects them, such as Deepgram or Google.

ElevenAgents runs on ElevenLabs’ own speech recognition, voices and turn-taking model, and you choose the LLM from a list that includes OpenAI, Anthropic and Google models, or connect your own. Its product page lists more than 16,000 voices and more than 90 languages, while its agent docs say an agent supports 31 languages when all are selected. Agents can switch language during a conversation once you add its language detection tool, which isn’t turned on automatically.

How does each keep answers current?

In Vapi, you upload files such as .txt, .pdf or .docx, select them on an assistant and publish, and Vapi creates a query tool that searches them. It suggests keeping each file under 300KB, and developers can replace the built-in search with their own retrieval server.

In ElevenAgents, documents come from files of up to 20MB, web pages, whole-site crawls or typed text. Small documents can sit in the prompt, while large ones use retrieval. Editing a document re-indexes it, URL documents can refresh on a schedule, and Google Drive files can sync.

Kawl Now takes a narrower route: FAQs, notes and selected GitHub docs, each approved before an agent can use it, with every GitHub re-sync sent back for approval.

Which handles phone calls and other channels?

Both take phone calls. Vapi gives you a free US number for inbound calls, and you import numbers from Twilio, Telnyx, DIDWW or a SIP trunk for outbound and international calls. It also runs outbound campaigns and can transfer a call to a person or another phone system.

ElevenAgents connects to Twilio and to SIP trunks from providers such as Telnyx, Plivo and Bandwidth, places batch outbound calls and can transfer a caller to a phone number. Beyond voice, the same agents can answer over WhatsApp, SMS, email and chat.

Kawl Now has no phone number for customers to call and doesn’t answer phone calls. Conversations happen in the visitor’s browser on your website. Human handover is an early release. A team verifies a UK or US phone number in Settings and turns handover on for its support or sales agent, and the person who answers hears a private briefing before the caller is connected. Customer routing is still being tested, so keep your usual contact options.

How do Vapi and ElevenLabs Agents charge?

Vapi starts with $5 of free credits, then charges a pre-paid $0.05 a minute for hosting, with transcriber, model and voice costs passed through. On 8 October 2026, its calculator estimated $82 to $129 a month for 1,000 minutes with Deepgram, OpenAI and ElevenLabs. Usage-only billing includes 4 concurrent calls. The Core package, at $29 a month, raises that to 10, and extra lines cost $10 a month each.

ElevenAgents’ free plan includes 15 minutes of calls and 4 concurrent calls. After that, pay-as-you-go and extra plan minutes cost $0.08 a minute, with LLM usage billed separately. Plans run from Starter at $6 a month for 75 minutes to Business at $990 a month for 12,375 minutes and 40 concurrent calls.

Kawl Now counts conversations, not minutes: each workspace gets its first 50 free, up to 10 minutes each. Paid calls aren’t available yet.

Which should you choose?

Choose Vapi if your developers want to pick each provider, see the cost of each part and build phone agents with transfers and outbound campaigns.

Choose ElevenAgents if voices, languages or channels matter most, or you’d rather use ElevenLabs’ own speech and voice models than pick each provider.

Choose Kawl Now if you run a SaaS website and want a support agent and a sales agent that answer visitors in the browser from sources your team approves, with your first 50 conversations free. It has no phone number for customers to call, no text chat and no language setting.

Whichever you shortlist, run the same test: give each the same docs, ask the same questions and listen to the answers.

Common questions