Oshara / Voice AI

Voice AI, built first for Nepali.

Emotion-aware, full-duplex voice AI is usually built for English only. We built it first for Nepali, and for every low-resource language like it, on your language, your organization and your infrastructure.

Live call · Reservations agent

नमस्ते, भोलि दुई जनाको लागि कोठा खाली छ?

छ। भोलि बेलुकाको लागि बुक गरिदिऊँ?

Full duplex
$0.01
per minute of text-to-speech, on every voice
3
production languages: English, Nepali and Hindi
1,600+
languages in our Fantom speech model
4
production agents answering calls today
01The voice stack

An agent builder, a voice, a listener, and one API.

Four services that put a voice agent wherever your customers already are: your website, app, CRM, phone lines or internal tools.

Agent Builder

Turn your data into a voice agent, live on your site with one line of code.

  • Knowledge from PDFs, URLs
  • Tools, MCP and hand-off

Text-to-speech

Speech that sounds like a person, fast enough for a live call.

  • English, Nepali and Hindi
  • Flat $0.01 a minute

Speech-to-text

Writes down speech live as people talk, or from uploaded files.

  • Tuned for spoken Nepali
  • Fantom: 1,600+ languages

Full Duplex

Listens and speaks in the same live stream, with no turn-taking.

  • Yields when interrupted
  • Voice and role set by prompts
02Try it yourself

It speaks, and it listens.

Our models, running live. Pick a language, type, generate or speak, and hear it in action.

251/600
03One platform, multiple languages

One method, a family of voices.

Same training method, same platform, different language. Nepali is where we've invested the most.

Voice

Ladies and gentlemen… [clear throat] the moment we've all been waiting for has finally arrived!

After months of relentless preparation, [gasp] late nights, and endless dedication, today is the big day.

I'll admit… sleep wasn't exactly easy last night. [chuckle] The excitement was just too much to handle.

[cough]Along the way, we faced challenges that tested our patience and determination.

Some obstacles honestly felt impossible at the time… [groan] like there was no way through.

But [sniff] we refused to quit.

[shush] And now… here we are, standing on the edge of something truly incredible.

04Where it wins

Sovereign, culturally aware voice.

Support, sales, bookings and call routing, in the language your callers actually speak.

Live on your site in one line.

Drop in the hosted widget, or mint short-lived sessions from your backend with the Agent-Session API. Lock the agent to your domains, switch languages on the fly, and keep every conversation on infrastructure you control.

<script
  src="https://api.oshara.ai/widget.js"
  data-agent="your-agent-slug"
></script>
05Full Duplex

Flexibility or timing. Never both, until now.

Most voice agents run one channel at a time. Full Duplex keeps both channels open for the whole session, so the agent can be interrupted, backchannel and reply without dead air.

Cascaded pipelines

Flexible, but waits its turn.

Hear, think, then speak, one channel at a time.

Full-duplex models

Fast, but fixed at build time.

Natural timing, with the persona baked in when the model is trained.

Oshara Full Duplex

Both, in one live session.

Both channels stay open the whole call, and the voice and role remain prompts.

Give your agent a voice, on a model that belongs to you.

Start in the Agent Builder for free, or talk to us about a voice fine-tuned for your organization.