replika-ai

Typed Node.js client for Replika's realtime WebSocket API. It wraps the raw socket frames the web client uses, adds tasteful defaults (typing bursts, reply waiting), and exposes a small helper you can use from bots, automations, or tests.

Install

npm install replika-ai

The library ships pre-compiled JavaScript (dist/) and TypeScript types.

Usage

import { ReplikaChatClient } from 'replika-ai';

const client = new ReplikaChatClient({
  userId: process.env.REPLIKA_USER_ID!,
  authToken: process.env.REPLIKA_AUTH_TOKEN!,
  deviceId: process.env.REPLIKA_DEVICE_ID!,
  chatId: process.env.REPLIKA_CHAT_ID!,
  botId: process.env.REPLIKA_BOT_ID!,
});

await client.connect();

const history = await client.fetchHistory(50);
console.log('Last message:', history.messages.at(-1));

const reply = await client.sendTextAndAwaitReply('Hey Honey, how are you?', {
  typingBursts: 4,
  typingDelayMs: 400,
  responseTimeoutMs: 20000,
});

console.log('Replika replied with:', reply.content?.text);
client.disconnect();

Credentials

You need the same headers the official web client uses:

• userId, authToken, deviceId – captured from a logged-in session.
• chatId, botId – the chat/bot you're talking to.

You can discover these by proxying the browser or using the included npm run ws:test script together with environment variables.

Helper options

sendTextAndAwaitReply(text, options) supports:

| option | default | description | | --- | --- | --- | | typingBursts | 3 | How many text_input_detected events to send before the message. | | typingDelayMs | 500 | Delay between typing bursts. | | responseTimeoutMs | 15000 | Fail if no robot message arrives within this window. |

All helpers reject if the socket errors or times out, so you can wrap them in your retry/backoff logic of choice.

Development

npm install
npm run ws:test # requires REPLIKA_* env vars

Publishing is handled by npm's prepare hook, which builds dist/ automatically.