Realtime Quickstart

This guide opens a Realtime session, sends one text message, and receives a spoken reply. Use the OpenAI SDK you already have, or a raw WebSocket.

Create an API key

Go to Settings → Developer → API Keys in app.60db.ai and create a key. Keep it on your server. The examples read it from SIXTYDB_API_KEY.

export SIXTYDB_API_KEY=sk_your_api_key

Never ship sk_ keys to a browser. For web pages, use browser ephemeral keys.

Open a session and send a text turn

JavaScript (openai)
Python (openai)
Raw WebSocket (Node)

import OpenAI from 'openai';
import { OpenAIRealtimeWS } from 'openai/realtime/ws';

const client = new OpenAI({
  apiKey: process.env.SIXTYDB_API_KEY,
  baseURL: 'https://api.60db.ai/v1',
});

const rt = await OpenAIRealtimeWS.create(client, { model: '60db-realtime-v1-mini' });

rt.send({
  type: 'session.update',
  session: {
    type: 'realtime',
    output_modalities: ['audio'],
    audio: { output: { voice: 'YOUR_VOICE_ID' } },
    instructions: 'You are a friendly assistant. Keep replies short.',
  },
});

rt.send({
  type: 'conversation.item.create',
  item: {
    type: 'message',
    role: 'user',
    content: [{ type: 'input_text', text: 'Say hello in one sentence.' }],
  },
});
rt.send({ type: 'response.create' });

rt.on('response.output_audio.delta', (event) => {
  // event.delta is base64 PCM16 at 24 kHz mono
});
rt.on('response.done', (event) => {
  console.log('usage:', event.response.usage);
  rt.close();
});

import asyncio
import os

from openai import AsyncOpenAI

client = AsyncOpenAI(
    api_key=os.environ["SIXTYDB_API_KEY"],
    base_url="https://api.60db.ai/v1",
)

async def main():
    async with client.realtime.connect(model="60db-realtime-v1-mini") as conn:
        await conn.session.update(session={
            "type": "realtime",
            "output_modalities": ["audio"],
            "audio": {"output": {"voice": "YOUR_VOICE_ID"}},
            "instructions": "You are a friendly assistant. Keep replies short.",
        })
        await conn.conversation.item.create(item={
            "type": "message",
            "role": "user",
            "content": [{"type": "input_text", "text": "Say hello in one sentence."}],
        })
        await conn.response.create()

        async for event in conn:
            if event.type == "response.output_audio.delta":
                pass  # event.delta is base64 PCM16 at 24 kHz mono
            if event.type == "response.done":
                print("usage:", event.response.usage)
                break

asyncio.run(main())

// npm install ws
import WebSocket from 'ws';

const ws = new WebSocket(
  'wss://api.60db.ai/v1/realtime?model=60db-realtime-v1-mini',
  { headers: { Authorization: `Bearer ${process.env.SIXTYDB_API_KEY}` } },
);

ws.on('open', () => {
  ws.send(JSON.stringify({
    type: 'session.update',
    session: {
      type: 'realtime',
      output_modalities: ['audio'],
      audio: { output: { voice: 'YOUR_VOICE_ID' } },
    },
  }));
  ws.send(JSON.stringify({
    type: 'conversation.item.create',
    item: {
      type: 'message',
      role: 'user',
      content: [{ type: 'input_text', text: 'Say hello in one sentence.' }],
    },
  }));
  ws.send(JSON.stringify({ type: 'response.create' }));
});

ws.on('message', (raw) => {
  const event = JSON.parse(raw);
  if (event.type === 'response.output_audio_transcript.delta') process.stdout.write(event.delta);
  if (event.type === 'response.done') ws.close();
});

Replace YOUR_VOICE_ID with a voice ID from List voices.

Send a voice turn

For voice, stream PCM16 audio into input_audio_buffer.append. With the default server_vad turn detection, the server commits the utterance after silence and starts the reply on its own.

{ "type": "input_audio_buffer.append", "audio": "<base64 PCM16 mono, 24 kHz>" }

Listen for these server events:

Event	What to do
`input_audio_buffer.speech_started`	Stop local playback (barge-in)
`conversation.item.input_audio_transcription.completed`	Show the user’s transcript
`response.output_audio.delta`	Play the audio chunk
`response.done`	Read `response.usage`

See Audio and turn detection for formats and manual turn control.

Check the result

The final response.done event carries usage. Each session is billed at $0.02 per connected minute plus LLM tokens. See Realtime overview.

Troubleshooting

Symptom	Cause	Fix
`401` or `authentication_error` on connect	Missing or wrong `Authorization` header	Send `Bearer sk_...` from a server
Close code `1008` on connect	Authentication failed	Check the key and that it is not deleted
Close code `4003`	Concurrent session limit reached	Close idle sessions or upgrade your plan
`invalid_value` on `session.audio.output.voice`	The voice is not available to your workspace	Use a stock voice or one of your own voices

Next steps

Migrate an existing OpenAI app
Function tools
60db extensions: memory, knowledge and intent

Get Started

Guides

Reference

Realtime Quickstart

Troubleshooting

Next steps

​Troubleshooting

​Next steps

Troubleshooting

Next steps