Docs GitHub

Talk to Any Website

A voice assistant your visitors can actually talk to. One line to install, no backend required.

Real-time Voice Page Understanding Interactive Highlights Open Source

Free tier, no credit card. Cloud or fully on-device. MIT-licensed.

Users Don't Read. They Scan.

Traditional chatbots add another text box to an already text-heavy page. The result? More reading, not less.

Text In, Text Out

Chat widgets force users to type questions and read answers. It's another wall of text on top of your existing content.

Voice In, Highlights Out

ArionTalk lets users speak naturally. The AI responds with voice while scrolling to and highlighting the exact content being discussed.

Beyond the Text Box

What makes ArionTalk different from every other chat widget.

Page Understanding

Automatically extracts text, images, and structure from any webpage. The AI knows what's on the page before you even ask.

Interactive Highlights

As the AI discusses content, it scrolls to and highlights the exact section or image, like a guided tour of your page.

Natural Voice with Barge-in

Talk naturally. Interrupt anytime. The AI stops, listens, and adapts.

Made for Content-Heavy Pages

Wherever visitors have to read a lot to find what they need.

Documentation sites

Visitors ask a question out loud; the AI reads them the exact section. Works alongside your existing search.

Product and pricing pages

Answer "does it integrate with X?" without making visitors scan a comparison table.

SaaS onboarding

Walk new signups through your getting-started guide in their own language.

Blogs and content sites

Let readers listen to articles, ask follow-up questions, and skip to the part they care about.

Try It Now

The widget is live on this page. Experience it in three steps.

1

Click the mic

Tap the microphone button in the bottom-right corner of this page.

2

Ask a question

Ask about anything on this page: features, engines, how it works.

3

Watch it highlight

See the AI scroll to and highlight the relevant content as it speaks.

From Voice to Highlight in Milliseconds

A real-time pipeline connecting your voice to page content.

User Speaks Web Audio API
Transcribe Speech Recognition
Gemini Live Multimodal API
Page Context + Images
Function Call highlight_and_scroll
Speak + Highlight Audio + Visual

Use Our Cloud. Or Run It Yourself.

Pick the hosted cloud service, run it on your own infrastructure, or keep everything fully offline.

Recommended

Cloud Engine

Powered by Gemini Live

  • 97 languages with automatic detection
  • Native audio streaming (low-latency)
  • 8 distinct AI voices
  • Interactive highlights
  • Image understanding
  • Works in any modern browser

Local Engine Developer preview

On-device, offline-first

  • Fully offline, no internet required
  • No server needed
  • Zero cost, completely free
  • Privacy-first, data stays in browser
  • Chrome 139+ only
  • English and Spanish only
  • Smaller model (Gemini Nano)

Want to run it on your own infrastructure? See the installation guide.

Start Free. Scale When You're Ready.

No credit card for the free tier. Move to Scale when you're ready.

Free

$0 / month

For solo devs and small teams getting started.

  • 50 voice minutes per month
  • 1 domain
  • All cloud engine features
  • Community support via GitHub Issues
Get Started Free

Scale

Talk to us

For teams with high traffic or custom needs.

  • Custom usage limits
  • Multiple domains
  • Custom voice and prompt
  • Dedicated support
Contact us

Add to Your Website in Seconds

One HTML tag. Voice ready out of the box. Customize with attributes.

index.html Get my snippet
<ariontalk-widget site-key="atk_live_..."></ariontalk-widget>

Common Questions

Does it work with my stack?

Yes. It's a Web Component, so it drops into any HTML page. That includes React, Vue, Svelte, Astro, and plain HTML.

What data does the AI see?

The widget sends the page's visible text, images, and the audio stream to Google's Gemini Live API. It doesn't set tracking cookies or profile visitors.

Which languages does it support?

The cloud service speaks 97 languages with automatic detection. The on-device engine supports English and Spanish.

What browsers does it work in?

Works in Chrome, Edge, Firefox, and Safari, on desktop and mobile. Microphone access is required.

Can I run it on my own infrastructure?

Yes. The whole project is open source under MIT. The installation guide has the details.

How much does it cost?

Free tier with no credit card. Paid plans for higher usage. See pricing above.

Ready to talk to your visitors?

One tag. Five minutes. No credit card.

Get Started Free