UG AI Engine for children
UG JS SDK 3.0
Last updated: 12 Oct 2025
🚀 UG-Core SDK Documentation
This document provides an overview of the
ug-core JS SDK, a framework-agnostic JavaScript library for managing real-time, interactive avatar conversations. It is designed to be decoupled from any specific UI framework, allowing for integration into various environments, such as React, Vue, or Vanilla JavaScript projects.Setup
Project repository → https://github.com/uglabs/ug-js-sdk
To set up your project locally - you can clone https://github.com/uglabs/ug-js-sdk
after cloning you can run
npm run serve:demo to try out the SDKHost: For Host URL, use: https://pug.stg.uglabs.app
API Key: To get a service account API key, go to https://pug-playground.stg.uglabs.app/service-accounts → click “create service account” → enter it’s name → copy the key
Federated ID: To get a player ID, go to https://pug-playground.stg.uglabs.app/players → click “create player” → enter player’s external id → copy federated id of the player you created (appears at the end of the list)
Install locally
npm install github:uglabs/ug-js-sdk
🏛️ Core Architecture
The SDK is built around a central
Conversation Manager that orchestrates several specialized managers to handle different aspects of the conversation lifecycle.- 🧞
Conversation Manager: The main entry point and central nervous system of the SDK. It initializes and coordinates all other managers, manages the overall conversation state (idle,playing,userSpeaking, etc.), and exposes a simple API for controlling the conversation.
- 🗣️
Conversation Network: Manages the WebSocket connection to the backend, handling the sending and receiving of all conversation-related messages.
- 🎩
User Input Manager: Captures user input. It uses aVAD Manager(Voice Activity Detection) to detect when the user is speaking and anAudioRecorderto capture microphone data. It can also handle text input.
- 🎪
Playback Manager: Manages the playback of the assistant's response. This includes playing audio (Audio Player), displaying subtitles (Subtitle Manager), and triggering avatar animations (Using Avatar Manager).
✨ Key Features
🗣️ Interrupts (Barge-in) [WIP]
The SDK is designed for natural, fluid conversations. Users can interrupt the assistant at any time, just like in a real conversation. This is achieved through robust Acoustic Echo Cancellation (AEC).
We leverage the browser's native AEC capabilities, which are often hardware-accelerated on modern mobile devices. In environments where hardware AEC is unavailable, a software-based solution is used, ensuring a seamless experience across all platforms.
We leverage the browser's native AEC capabilities, which are often hardware-accelerated on modern mobile devices. In environments where hardware AEC is unavailable, a software-based solution is used, ensuring a seamless experience across all platforms.
🤫 Smart Turn Detection [WIP]
No awkward silences or waiting for a push-to-talk button. The SDK uses Smart Turn Detection Model to intelligently detect when a user has started and, more importantly, finished speaking. This allows for a natural conversational flow, where the assistant responds precisely when the user has completed their thought.