Persona Designer
Tone, personality traits, and greeting style - every conversation sounds like your brand.
Oshara Agent Builder
Build Your Own
Voice AI Agent
Deploy It Anywhere
Turn your data into a sovereign voice agent - persona, knowledge, tools, and voice, live on your site in one line of code.
Text-to-speech $0.01 / minPersona & knowledgeTools & MCP serversOne-line embedNo ML team required
Powered & Supported by
Live on these sites today
Real businesses running Oshara agents on live traffic. Click any one and talk to it yourself.
4production agents
answering calls right now
answering calls right now
From Idea to Live Agent in Five Steps
1
Shape the Persona
Name it, describe it, set personality and greeting - no code.
2
Add Knowledge & Tools
Upload documents, connect URLs and tools so your agent can answer and act.
3
Pick Your Models
Choose the LLM, our Lyric or Fantom speech models, and the TTS voice that powers your agent.
4
Customize Appearance
Set the widget title, brand colors, avatar, and display settings - previewed live as you edit.
5
Deploy Anywhere
Drop one script tag on your site, or call the agent-session API. The widget appears automatically.
A Full Studio for Agentic Voice
Persona, knowledge, tools, models, appearance, and forms - tuned in one place, with a live preview of your widget the whole way.
Knowledge Base
PDFs and URLs, chunked and retrieved automatically mid-conversation.
Tools & MCP Servers
MCP servers, HTTP endpoints, and client events - your agent can act, not just talk.
Model Control
Any LLM, our Lyric (90+ languages) or Fantom (1600+ languages) speech models, and a TTS voice library.
Branded Appearance
Name, colors, avatar, and watermark - previewed live as you edit.
Multi-Step Forms
Forms your agent surfaces mid-call to book demos, capture leads, or run intake.
Pick a Voice That Fits Your Brand
Production voices tuned for English, Nepali, and Hindi - preview each one before you assign it to an agent.
Built for Real-World Constraints
Sovereign, low-latency, and fine-tuned on your data - the infrastructure behind every agent.
Sovereign Security
All audio and data processed locally - nothing sent to foreign clouds.
Low Latency
Optimized for real-time voice UX, production-ready out of the box.
Language & Dialect Precision
Fine-tuned on Nepali and regional dialects - nuance, grammar, and vocabulary included.
Sentiment-Adaptive Tone
The voice model adapts tone to emotional context, reducing churn.
Fine-Tuning Automation
OFTA handles data prep, training, and evaluation - bring your own data, we run the pipeline.
Hand Off Conversations Between Agents
Wire up call-transfer rules so a conversation moves - with full context - to the right specialist.
Call Transfer Settings
Decide when a conversation escalates, then preview the hand-off.
Allow agent-to-agent transfer
Carry conversation context
Auto-escalate on intent match
Support Bot
First responder
Billing Agent
Payments & invoicing
Tech Support
Troubleshooting
Live Human
Senior escalation
One agent routes a live conversation out to the right specialist.
Real Conversation, Both Ways at Once
Your agent listens and speaks in the same live session - voice and persona set independently, so it never waits its turn.
Turn-based todayTypical agents
Caller
Full Duplex
One channel at a time. Fast, but each side still waits for a turn.
Full-duplexOshara
Caller
Full Duplex
Both channels open. Speech overlaps, as it does with people.
AgentUser
Restaurant Reservations
EnglishLoan Repayment Reminder
NepaliGive Your Agent a Form to Work With
Pick a template or build your own — your agent surfaces the right form mid-call, fields pre-filled from the conversation.
General Inquiry
Get in Touch
I'll make sure the right person sees this right away.
Name
Sam Rivera
Email
sam@email.com
Phone number
+1 (555) 204-8821
Team size
11–50
Message
Describe your question…
Choose a template
Ready-made forms,
live in the widget.
Click a template to preview it - your agent picks automatically mid-call.
Forms that fill themselves - detected mid-call, pre-filled, ready to confirm.
Your Agent. Your Website. Two Ways to Ship.
Embed the hosted widget with one snippet, or build a fully custom experience on top of the agent-session API.
No code
Hosted Widget
We host it - paste one script tag on any page and the widget appears.
<script
src="https://api.oshara.ai/widget.js"
data-agent="your-agent-id">
</script>Code
Agent-Session API
Mint a session with your API key and connect your own UI to the live agent.
curl -X POST https://api.oshara.ai/api/agents/agent-session/ \
-H "x-api-key: sk_..." \
-H "Content-Type: application/json" \
-d '{ "agent": "your-agent-id" }'Lock it to your domains. Switch languages on the fly.
Every conversation stays on infrastructure you control.
One Cent a Minute of Speech
You pay for the audio your agent actually generates - a flat $0.01 per minute of text-to-speech. No seats, no platform fee, no idle-time billing.
Text-to-speech
$0.01/ min
That's $0.60 an hour of generated speech - the same rate whether you ship one agent or a hundred.
- No seats, no platform fee
- Same flat rate on every voice
Estimate your bill100 min
$1.00total
100 min × $0.01 per minute
≈ 1.7 hours of generated audio
100 min100k min
Final usage is billed per minute generated.
Ready to Build Your Agent?
Design the persona, feed it your knowledge, and deploy to your site in an afternoon - no ML team required.