Skip to main content
60db Realtime is a speech-to-speech API over a single WebSocket. You stream microphone audio in and get spoken replies back, with server-side turn detection, barge-in and function calling. It speaks the OpenAI Realtime GA protocol, so the OpenAI SDKs and @openai/agents-realtime work unchanged once you point them at 60db. Migrating from OpenAI? Change three values: the URL, the model and the key. See Migrate from the OpenAI SDK.

OpenAI-compatible

Same events, same session shape. Unmodified OpenAI clients connect to wss://api.60db.ai/v1/realtime

One socket

Speech recognition, the LLM turn and voice synthesis run on the server. No stitching three APIs together

60db voices

Use any 60db voice, including your own cloned or professional voices in the same session

Browser-safe keys

Mint short-lived ek_ keys on your server and connect from the browser without exposing sk_

​
Endpoints

Details: WebSocket connection and Create ephemeral key.

​
Models

A larger model tier is planned but not available yet. Use 60db-realtime-v1-mini today.

​
Pricing

Realtime is billed per connected minute, plus LLM tokens. There is no separate charge for speech-to-text or voice.
  • Billed in 60-second ticks and prorated per second when the session closes.
  • Charged to your workspace balance. See Plans, Credits & Top-ups.
  • A 10-minute call costs about $0.20 plus tokens.

​
Limits

Idle and maximum-duration values are enforced by the server and may differ by plan. Handle close codes 4000 and 4001 by reconnecting or ending the call.

​
Supported and not yet supported

​
Next steps

Quickstart

Run a text turn and an audio turn in a few minutes

Migrate from OpenAI

Change the URL, model and key in your existing app

Browser ephemeral keys

Connect from a web page safely

API reference

Connection, client events and server events