Overview
Intro to Realtime API (Speech-to-Speech)
Inworld's Realtime API (Speech-to-Speech) enables low-latency, speech-to-speech interactions, with an optimized STT, LLM, and TTS pipeline that delivers the best latency, quality, and flexibility.
The API follows the OpenAI Realtime protocol, extended to enable additional customization.
WebSocket Quickstart
Build a voice agent with WebSocket, mic input, and audio playback.
WebRTC Quickstart
Build a voice agent with browser-native WebRTC — no manual audio encoding.
API reference
See the full event schemas for the Realtime API.
JS examples
JavaScript examples for the Realtime API.
Python examples
Python examples for the Realtime API.
Using AI to code? Paste https://docs.inworld.ai/llms.txt into your assistant so it knows every page on this site. Want live search? Add the MCP server.
Prefer the terminal? npm install -g @inworld/cli — create and manage API keys for the Realtime API with the Inworld CLI. AI agents can use it too.
Key Features
- WebSocket and WebRTC transports: Connect over WebSocket or WebRTC with a standard event schema.
- Automatic interruption-handling and turn-taking: Your agent will manage conversations naturally and be resilient to user barge-in.
- Conversational awareness: With Realtime TTS-2, the model conditions on the audio of prior conversational turns. A line delivered after a joke lands differently than the same line delivered after bad news. The model hears the difference and adjusts how it speaks based on how it was spoken to.
- Router support: Utilize Realtime Router to enable a single agent to dynamically handle different user cohorts, or to facilitate A/B tests.
- OpenAI compatibility: Drop-in replacement for the OpenAI Realtime API with a simple migration path.
Guides
Configuring models
Configure sessions, send input, and orchestrate responses.
Managing conversations
Session lifecycle and conversation events.
OpenAI migration
Step-by-step guide to switch from OpenAI to Inworld.
See the API reference for full event schemas.