Skip to main content

​
The voice pipeline

Every conversation on Orato flows through a real-time pipeline. From when a caller speaks to when they hear a response, the entire loop completes in under one second.
Orato Voice AI Pipeline — Transcriber, LLM, Voice Engine
1

Transcriber — your assistant's ears

The moment someone speaks, Orato transcribes their voice into text in real time using Automatic Speech Recognition (ASR). The AI cannot process audio directly; it needs text. A fast, accurate transcriber ensures the AI understands the caller correctly every time.
2

LLM — your assistant's brain

The transcribed text is sent to a Large Language Model. The LLM reads the full conversation history, follows your written instructions, and generates a reply. This makes your assistant intelligent. It understands context, handles objections, and makes decisions mid-call.Orato supports three LLM providers:
  • Anthropic: Claude models, nuanced and careful reasoning
  • Groq: Llama 3.3 70B, ultra-fast processing for low-latency calls
  • OpenAI: GPT-4 models, balanced quality and speed
3

Voice Engine — your assistant's voice

The LLM’s text response is converted into natural speech using Text-to-Speech (TTS). The caller hears a real voice, not a robot. Orato supports:
  • Cartesia: sonic-2, low-latency and natural (recommended for most)
  • ElevenLabs: highly natural voices, wide selection
  • Sarvam: Hindi, Hinglish, and Indian English accents

​
Key concepts

An AI Assistant is a fully configured voice agent. Give it a name, write its instructions, and choose its LLM and voice — and it handles calls on your behalf. One assistant can run unlimited simultaneous conversations.Each assistant shows a per-minute cost (~0.09/min) that combines Transcriber + LLM + Voice + Telephony + Platform costs.
Instructions are what you write to tell your assistant how to behave on every call. They define:
  • Who the assistant is (name, company, role)
  • What the goal of the call
  • What questions to ask
  • What to avoid saying
  • How to end the conversation
Orato’s AI Instruction Generator auto-generates a structured prompt — just describe your business and call goal, then click Generate.
A Lead Batch is a group of contacts you want to call. You create batches manually or by uploading a CSV file. Batches then serve as the target list when creating a Campaign.
A Campaign connects a Lead Batch to an AI Assistant and automatically calls every contact. You set the call rate (3–8 calls per minute recommended), apply lead filters, and schedule when to start. Campaign progress is tracked in real time.
Every voice call and web chat your assistant handles is logged as a Conversation — including contact details, duration, an AI-generated summary, outcome, and full transcript. Voice and chat conversations are tracked separately.
Upload PDF, TXT, or DOCX files (up to 10MB each), and your assistant can search and reference them during conversations. Useful for product FAQs, pricing sheets, policies, and scripts.
A floating microphone button that visitors click to talk to your AI Assistant in real time — directly on your website. Configured with your brand colour, logo, and position. Added via a single script tag.
Phone numbers are what your assistants use to make and receive calls. You buy numbers directly from Orato (US and India) after completing a one-time compliance request with your business documents.

​
Cost breakdown

Every AI Assistant shows a per-minute cost bar in the dashboard:
ComponentWhat it covers
TranscriberSpeech-to-text conversion cost
LLMLanguage model inference (varies by model and provider)
VoiceText-to-speech synthesis
TelephonyPhone call infrastructure (~$0.02/min)
PlatformOrato platform fee
A typical assistant configuration costs approximately 0.09 per minute.

​
Assistant roles

RoleHow it’s deployed
Web WidgetEmbedded voice/chat assistant on your website
OutboundMakes calls to leads via Campaigns
InboundAnswers calls to your Orato phone number.

​
Next steps

Quick Start

Create your first assistant and go live

AI Assistants

Full guide to building and configuring assistants