Skip to main content
Start with the job you have today. You can add the other paths later. Every path uses the same key and the same models.
90 seconds to a transcript. Mint a sandbox key, then transcribe a file:
Expected: { "text": "..." }. That path uses scopes the sandbox mint already has. Clone does not; check GET /v1/me first. Full copy-paste paths: Quickstart.

Use speech APIs

Hear, Speak, Clone, Cast or Dub. Mint a sandbox key and make a first call.

Build a voice agent

Build through the console or API, test with Omni, then connect your website, phone or application.

Keep your framework

LiveKit or Pipecat owns the loop. PyAI is the speech layer: pyai.STT and pyai.TTS.

​
API primitives

Use this when you already own transport, the LLM, or turn policy.
  1. Mint a sandbox key (POST /v1/sandbox/keys).
  2. Choose Speech To Text (Hear) for transcription, Text To Speech (Speak) for synthesis, or Clone for a custom voice.
  3. Open Omni yourself with the browser voice agent guide when you want the full loop in code.
Authentication covers bearer keys, scopes, and the sandbox mint. SDKs covers pip install pyai-sdk and npm install @pyai/sdk. Pricing lists current rates and included usage. Before launch, complete the production-readiness checklist.

​
Dub an existing recording

Use Dub to turn an English recording into Hindi audio. Submit the source, poll the job, then download its WAV output. Hindi is the currently enabled target; this is an asynchronous workflow.

​
Agents through UI or API

Use Agents for a complete voice conversation, powered by the Omni realtime runtime. Choose the builder or API to manage the same saved profile.
  1. Build an Agent in the console or create one through the API.
  2. Add a hosted knowledge base so answers come from your pages and files. Paste a site URL and PyAI crawls the public pages.
  3. Publish the website Call Now button, or assign a number.
Creating the same profile from code is Create agents via API.

​
Framework adapters

Use this when LiveKit Agents or Pipecat should keep the session, tools, and LLM. These adapters use Hear and Speak. They do not create or select an Omni agent.

​
Next

Get an instant key

Mint a sandbox key without signup, email, card, or billing.

SDKs

pip install pyai-sdk, npm install @pyai/sdk, plus LiveKit and Pipecat.

Omni overview

One WebSocket for speech, reasoning, tools, and knowledge.

Errors and limits

First-call failures, stable codes, and when not to retry.