Skip to main content

​
The voice pipeline

Every conversation on Orato flows through three stages that repeat for each turn: the transcriber listens, the AI model thinks, and the voice engine speaks.
Orato voice pipeline: transcriber, AI model, voice engine
1

Transcriber: your assistant's ears

Converts the caller’s speech into text in real time. Orato picks a transcriber automatically based on the languages you choose. You can override it in Advanced settings with Deepgram (Flux English, Flux Multilingual) or Sarvam AI (Saaras v4 Realtime, Saaras v3).
2

AI model: your assistant's brain

A large language model reads your instructions, the knowledge you attached, and the conversation history, then writes the next reply. Providers: Anthropic, Groq, OpenAI, Sarvam AI, and Google. See Providers.
3

Voice engine: your assistant's voice

Converts the reply into speech. Providers: Cartesia, ElevenLabs, Sarvam, and Smallest AI, each with its own voices and speed range.

​
Key concepts

A configured voice agent. It has a name, a first message, instructions, a voice, an AI model, and call settings. One assistant can handle many conversations. See AI Assistants.
Every assistant has one role that decides how it is deployed:
The written guide your assistant follows on every conversation: who it is, what the goal is, how to speak, and what to avoid. Orato can generate instructions for you and structure them into sections. See Prompt.
Placeholders such as {{customer_name}} that Orato fills in for each lead, so the assistant can greet people by name.
A named group of contacts, created manually or from a CSV or Excel file. Campaigns call one batch at a time. See Leads.
Connects a lead batch to an outbound assistant and calls every lead at the call rate you set, immediately or on a schedule. See Campaigns.
A record of one voice or chat interaction, with contact, assistant, duration, summary, outcome, and billing. See Conversations.
Documents (PDF, TXT, DOCX) and imported website pages your assistant searches during calls and widget chats. See Knowledge base.
Your prepaid USD balance. Calls and phone numbers are paid from the wallet. See Billing and credits.

​
Cost per minute

While you build an assistant, Orato shows a live per-minute estimate at the top of the editor, in USD and INR, broken down by component: The estimate updates as you change providers, so you can compare options before saving. For example, a default setup with Anthropic Claude Haiku 4.5 and a Cartesia voice shows about $0.059 per minute. Phone number rental is billed separately (see Phone numbers).
Figures in this documentation reflect the dashboard at the time of writing. Always check the live estimate in the assistant editor for current pricing.

​
Next steps

Quick start

Create your first assistant.

AI Assistants

Full guide to every assistant setting.