Voice

How it works

The mic button is built into the widget — nothing to enable. A customer taps it, speaks naturally, and the bot detects when they've stopped, transcribes, retrieves from your knowledge base, and replies out loud. They can interrupt mid-answer, just like a phone call.

Languages

Voice runs on speech models built specifically for Indic languages, so Hindi, Tamil, or Telugu sound natural rather than translated. Voice is live today in English and 11 Indian languages: Hindi, Tamil, Telugu, Kannada, Malayalam, Marathi, Bengali, Gujarati, Punjabi, Odia, and Urdu.

Text chat additionally works in Chinese, Malay, Japanese, Thai, Vietnamese, and Indonesian. Voice for those is on the roadmap.

Requirements

• Your site must be served over HTTPS — browsers block microphone access otherwise.
• Embedding inline? Keep allow="microphone" on the iframe:

<iframe src="https://mongpt-api.mongpt.workers.dev/embed/YOUR_TENANT_ID"
  allow="microphone" width="100%" height="600"></iframe>

The customer's browser will ask their permission the first time. If they decline, chat still works normally.

Voice minutes

Voice is metered separately from messages: 60 minutes on Starter, 180 on Growth, 600 on Scale. Minutes reset monthly and don't roll over. Track usage on the Overview page. See Plans & quotas.

Multi-bot

Voice conversations are scoped to the bot in the embed — a customer talking to your storefront bot gets answers from that bot's knowledge base, never another's.

Tips

• Voice answers are read aloud, so keep your fallback message short — long fallbacks sound worse spoken than written.
• Set the bot language to the one most of your callers use; the bot will still follow a customer who switches.
• Test in the Playground before you ship.