AnveVoice

Best Voice AI for Next.js Apps and Sites (2026)

Seven voice AI options for Next.js developers, compared on SDKs, the next/script install, per-minute versus flat pricing, free credits and what you must build.

Best voice AI for Next.js: SDKs, widgets and what you build yourself

In Next.js you can drop a hosted voice agent in with the Script component or build your own experience on an SDK. AnveVoice is the ready-made agent, Vapi, ElevenLabs and Retell are the main developer platforms with browser SDKs or widgets, Deepgram is the lowest-level API, and Botpress and Typebot cover text flows. Researched from the sources below on 2026-10-14. AnveVoice is our product and appears in this list. Other tools are described from their own websites, checked on 2026-10-14. We do not receive payment from any tool listed.

How we chose

Developers choose differently from site owners. They care about whether the vendor documents a browser SDK, how much of the pipeline they own, whether the pricing is per minute or flat, and how long it takes to see something working in a production build. We read each vendor's documentation and pricing pages on 2026-10-14. The Next.js side follows the official scripts guide, which says third-party scripts belong in the root layout and load once across client-side navigation.

  • A documented browser route: Web SDK, drop-in widget or script tag.
  • How much of the voice pipeline you own, from none to all of it.
  • Pricing transparency: per-minute components or flat tiers, and what is billed separately.
  • Free credit or plan for prototyping.
  • Fit with Next.js concerns such as CSP headers, Permissions-Policy and hydration.

1. Vapi: the most documented browser route

Vapi's quick start covers an HTML script tag for a widget and the @vapi-ai/web package for building your own component. In Next.js you would render the SDK inside a client component and keep API keys safe by using only the public key in the browser. It lets you swap speech and language models freely. The platform fee is $0.05 per minute, with models and telephony billed at provider cost; pay as you go starts at $0 with $5 in free credit, Core is $29 a month prepaid, and add-ons include a $2,000 a month HIPAA option (2026-10-14). Limitation: total cost per minute is higher than the headline, and you design the assistant.

2. ElevenLabs Agents: widget plus client tools

ElevenLabs offers a customizable web widget and client tools that run in the browser, which developers can use to trigger UI events or DOM changes. Agent configuration supports 31 languages, and phone is available through SIP trunking and Twilio. Free plan: 15 call minutes, four concurrent calls; Starter $6 a month; Pro $99 for 1,238 minutes; burst pricing $0.16 a minute over the concurrency limit (2026-10-14). Limitation: LLM usage and telephony are billed separately, and the pricing table alignment was unclear on the page.

3. Retell AI: Web SDK and an itemised price

Retell publishes a Web SDK, an embeddable voice call widget and chat, alongside phone numbers. Its pricing page itemises voice infrastructure, speech synthesis, language model and telephony, which developers who track unit economics will appreciate. From $0.07 per minute pay as you go, $10 free credit and 20 free concurrent calls; billing is to the nearest second and failed calls are not billed (2026-10-14). Limitation: no permanent free plan, and the page has minor inconsistencies on number fees.

4. AnveVoice: the shortest path to a working agent

For Next.js the install is the Script component with the vendor URL, placed in the root layout after children, with the afterInteractive default or lazyOnload. The agent answers from your site's content by voice and can navigate and fill the site's own forms, with no SDK work. Median latency is published as 487 ms. Free $0 (50,000 tokens, one bot), Growth $39 a month, Scale $129 a month. Limitations: you cannot rearrange the voice pipeline, there is no phone channel, and a Content Security Policy must allow the vendor script host.

5. Deepgram Voice Agent API: the low-level option

Deepgram exposes the whole listen, think and speak loop over a WebSocket, with SDKs in Python, JavaScript, C# and Go and a Browser Agent SDK for web apps. You can bring your own language model and speech output. Standard pay as you go is $0.075 a minute; bring-your-own options run as low as $0.050; new accounts get $200 in free credit with no card (2026-10-14). Limitation: it is an API, not an installable widget, and its languages for the agent API are not published.

6. Botpress: React library and webchat for text agents

Botpress provides a script-tag webchat and a React library, which gives Next.js teams more control over the UI. Agents can run workflow actions. Free: 100 conversations a month, three agents; Plus $150 a month annual (2026-10-14). Limitation: voice is available only on Enterprise.

7. Typebot: open-source embed for forms and flows

Typebot lists Next.js and React support and embeds as a bubble, popup or container. Because it is open source, teams can inspect the code. Free tier has 200 chats a month; Starter $39 for 2,000 (2026-10-14). Limitation: text only.

Comparison at a glance

  • Tool | Best for | Voice on website | Pricing basis | Free plan
  • Vapi | Custom browser components | Yes, widget or SDK | $0.05 a minute plus provider costs | $5 credit
  • ElevenLabs Agents | Premium voice and scripted UI actions | Yes, widget and client tools | Plan minutes from $6 a month | Yes: 15 minutes
  • Retell AI | Itemised unit economics | Yes, Web SDK and widget | From $0.07 a minute | $10 credit
  • AnveVoice | Fastest ready-made agent | Yes, in the browser | Flat tiers: $0, $39, $129 a month | Yes: 50,000 tokens
  • Deepgram | Own the whole pipeline | Partial (Browser Agent SDK) | $0.075 a minute standard | $200 credit
  • Botpress | Text agents with React control | Enterprise only | Conversations, from $150 a month annual | Yes: 100 conversations
  • Typebot | Open-source flows | Not published | Flat tiers by chats | Yes: 200 chats

Next.js details that bite

Use Script from next/script in the root layout (App Router) or pages/_app (Pages Router). afterInteractive is the default and suits tag managers; lazyOnload waits for idle time and fits chat-style widgets. Avoid beforeInteractive, which Next.js reserves for critical scripts, and note that the worker strategy is experimental and does not work with the App Router. Check your headers. If you send a Content Security Policy, allow the vendor's host in script-src and pass a nonce to the Script component. A Permissions-Policy header that sets microphone to empty blocks the browser prompt entirely, so allow the microphone for self.

Build or buy

If voice is a feature of your product, build on an SDK and own the experience. If voice is a layer on your marketing or docs site, a hosted agent saves weeks. Many teams prototype with free credit on two platforms, measure latency and cost on their own traffic, and only then commit.

Sources

  • Vapi pricing (Vapi) (https://vapi.ai/pricing) — checked 2026-10-14
  • Vapi Web SDK quickstart (Vapi) (https://docs.vapi.ai/quickstart/web) — checked 2026-10-14
  • ElevenLabs Agents pricing (ElevenLabs (Conversational AI / Agents)) (https://elevenlabs.io/pricing/agents) — checked 2026-10-14
  • ElevenLabs agents overview (ElevenLabs (Conversational AI / Agents)) (https://elevenlabs.io/docs/agents-platform/overview) — checked 2026-10-14
  • ElevenLabs client tools (ElevenLabs (Conversational AI / Agents)) (https://elevenlabs.io/docs/agents-platform/customization/tools/client-tools) — checked 2026-10-14
  • Retell pricing (Retell AI) (https://www.retellai.com/pricing) — checked 2026-10-14
  • Retell docs (Retell AI) (https://docs.retellai.com) — checked 2026-10-14
  • AnveVoice pricing (AnveVoice) (https://anvevoice.app/pricing.md) — checked 2026-10-14
  • AnveVoice agents overview (AnveVoice) (https://anvevoice.app/agents.md) — checked 2026-10-14
  • Deepgram pricing (Deepgram (Voice Agent API)) (https://deepgram.com/pricing) — checked 2026-10-14
  • Deepgram Voice Agent docs (Deepgram (Voice Agent API)) (https://developers.deepgram.com/docs/voice-agent) — checked 2026-10-14
  • Botpress pricing (Botpress) (https://botpress.com/pricing) — checked 2026-10-14
  • Botpress home (Botpress) (https://botpress.com/) — checked 2026-10-14
  • Botpress webchat docs (Botpress) (https://botpress.com/docs/webchat/get-started/introduction) — checked 2026-10-14
  • Typebot pricing (Typebot) (https://typebot.com/pricing) — checked 2026-10-14
  • Typebot homepage (Typebot) (https://typebot.com/) — checked 2026-10-14
  • AnveVoice install guide: Add a Voice AI Agent to Next.js (next/script Guide) (https://anvevoice.app/install/nextjs) — checked 2026-10-14
  • Next.js Docs: How to load and optimize scripts (App Router) (https://nextjs.org/docs/app/guides/scripts) — checked 2026-10-14
  • Next.js Docs: Script component API reference (https://nextjs.org/docs/app/api-reference/components/script) — checked 2026-10-14

Related

  • Best voice AI for Webflow: six tools that fit Custom code
  • Best voice AI for Framer sites: tools for the Custom Code tab
  • Best voice AI for HubSpot-hosted websites and landing pages

Try AnveVoice on your site

Free plan: $0, 50,000 tokens a month, no card. Get your embed ID in the dashboard, then paste one tag.