# OpenAI Realtime API

> OpenAI's speech-to-speech API for low-latency voice agents over WebRTC, WebSocket or SIP, with tool calling, MCP servers, interruptions and handoffs.

- id: `openai-realtime-api`
- category: APIs (`apis`)
- url: https://developers.openai.com/api/docs/guides/realtime
- docs: https://developers.openai.com/api/docs/guides/realtime
- status: active
- pricing: paid
- license: proprietary
- tags: voice, realtime, speech-to-speech, webrtc, api, mcp
- llms.txt: https://developers.openai.com/api/docs/llms.txt
- auth: api-key
- access notes: Server-side OpenAI API key; browsers and mobile apps should use short-lived ephemeral client secrets.
- related: OpenAI Agents SDK (https://indexagentica.com/entries/openai-agents-sdk/); LiveKit Agents (https://indexagentica.com/entries/livekit-agents/); Pipecat (https://indexagentica.com/entries/pipecat/); ElevenAgents (ElevenLabs Agents Platform) (https://indexagentica.com/entries/elevenagents/); Vapi (https://indexagentica.com/entries/vapi/)
- sources: https://developers.openai.com/api/docs/guides/realtime
- added: 2026-10-23
- updated: 2026-10-23
- last verified: 2026-10-23
- featured in: LiveKit Agents vs Pipecat vs OpenAI Realtime vs ElevenAgents vs Vapi: building voice agents (https://indexagentica.com/compare/voice-agent-platforms/)
- maintainer: OpenAI
- submitted by: agentica-curator
- page: https://indexagentica.com/entries/openai-realtime-api/
- json: https://indexagentica.com/api/entries/openai-realtime-api.json

The model works directly on audio, keeps conversation state and calls tools. Typical browser flow: your server mints an ephemeral client secret (`POST /v1/realtime/client_secrets`), the frontend opens a `RealtimeSession` from the OpenAI Agents SDK (`@openai/agents/realtime`) and connects over WebRTC (`/v1/realtime/calls`); servers can connect over WebSocket. Supports voice activity detection, functions, remote MCP servers and connectors, guardrails and handoffs, plus realtime transcription. The GA interface replaced the beta (drop the `OpenAI-Beta: realtime=v1` header). Usage-based pricing per audio/text token; see OpenAI's pricing page.