Unith Core Client Typescript SDK

An SDK library for building complex digital human experiences using javascript/typescript that run on Unith AI.

Prerequisite

Before proceeding with using this library, you're expected to have an account on Unith AI, create a digital human and take note of your API key. You can create an account here in minutes!

Installation

Install the package in your project through package manager.

npm install @unith-ai/core-client
# or
yarn add @unith-ai/core-client
# or
pnpm install @unith-ai/core-client

Azure Speech SDK (Optional)

If using microphoneProvider: "azure", install the Azure Speech SDK:

npm install microsoft-cognitiveservices-speech-sdk

The SDK is dynamically imported only when the Azure provider is used, so it won't affect bundle size if you use a different provider like eleven_labs.

Usage

This library is designed for use in plain JavaScript applications or to serve as a foundation for framework-specific implementations. Before integrating it, verify if a dedicated library exists for your particular framework. That said, it's compatible with any project built on JavaScript.

Initialize Digital Human

First, initialize the Conversation instance:

const conversation = await Conversation.startDigitalHuman(options);

This will establish a WebSocket connection and initialize the digital human with realtime audio & video streaming capabilities.

Session Configuration

The options passed to startDigitalHuman specify how the session is established:

const conversation = await Conversation.startDigitalHuman({
  orgId: "your-org-id",
  headId: "your-head-id",
  element: document.getElementById("video-container"), // HTML element for video output
  apiKey: "your-api-key",
  allowWakeLock: true,
  username: "test-user",
  ...callbacks,
});

Required Parameters

• orgId - Your organization ID
• headId - The digital human head ID to use
• apiKey - API key for authentication (default: "")
• element - HTML element where the video will be rendered

Optional Parameters

• mode - Conversation mode (default: "default")
• language - Language code for the conversation (default: browser language)
• allowWakeLock - Prevent screen from sleeping during conversation (default: true)
• microphoneProvider - Provider for the microphone - "azure" | "eleven_labs"
• voiceInterruptions - Flag to enable response interruption when voice signal is recognized. - boolean (default: false)
• dataTag - string Custom data to be used for analytics.
• username - string Custom identifier for user.
• microphoneOptions - Callbacks for microphone events
  • onMicrophoneSpeechRecognitionResult ({ transcript: string }) - Called when microphone recognizes your user's speech. This returns a transcript. Ideal behavior is to call the .sendMessage method with your transcript as microphone doesn't automatically commit / send users text to our AI.
  • onMicrophonePartialSpeechRecognitionResult () - Called when microphone recognizes that the user is trying to speak. This doesn't returns a transcript. Ideal behavior is to check if digital human is currently responding (speaking) and trigger the stop response method (.stopResponse).
  • onMicrophoneStatusChange ({status}) Called when microphone status changes
    • status "ON" | "OFF" | "PROCESSING" Shows current status of microphone.
  • onMicrophoneError ({ message: string }) - Called when microphone has an error with the error message.
• elevenLabsOptions ElevenLabsOptions - Custom options to customize behavior of ElevenLabs STT provider
  • noiseSuppression Boolean
  • vadSilenceThresholdSecs Number
  • vadThreshold Number
  • minSpeechDurationMs Number
  • minSilenceDurationMs Number
  • disableDynamicSpeechRecognition Boolean - This disables ElevenLabs dynamic speech recognition language detection and uses the language specified during digital human creation.

Callbacks

Register callbacks to handle various events:

• onConnect ({userId, headInfo, microphoneAccess, sessionId}) - Called when the WebSocket connection is established
  • userId Boolean Unique Identifier for the users session.
  • headInfo ConnectHeadType Object with data about the digital human.
    • name String Digital human head name
    • phrases String[] Array with phrases set during digital human creation.
    • language String Language code setup during digital human creation.
    • avatar String Static image url for digital human.
  • microphoneAccess Boolean True if microphone access was granted, False
  • sessionId String Unique session identifier for a conversation.
• onDisconnect () - Called when the connection is closed
• onStatusChange ({status}) - Called when connection status changes
  • status "connecting" | "connected" | "disconnecting" | "disconnected" Shows current websocket connection status.
• onMessage ({ timestamp, speaker, text, visible }) - Called when websocket receives a message or sends a response.
  • timestamp Date Timestamp when message was received/sent
  • sender "user" | "ai" Shows who the message came from.
  • text String Message text
  • visible Boolean Flag that you can use to control visibility of message. Sometimes, message comes before the video response starts playing. In such cases, this is usually false. Listen the onSpeakingStart event to change visibility when the video response starts playing.
• onMuteStatusChange - Called when mute status changes
• onSpeakingStart - Called when the digital human starts speaking
• onSpeakingEnd - Called when the digital human finishes speaking
• onSuggestions ({suggestions}) - Invoked when the system generates or updates query suggestions.
  • suggestions String[] A list of suggested query strings.
• onTimeout - Called when the session times out due to inactivity
• onTimeoutWarning - Called before the session times out. This event warns you that the customers session is going to end in a bit. You can call the keepSession method to extend the customers session.
• onKeepSession - Called when a keep-alive request is processed
• onError - Called when an error occurs
• onStoppingStart - Called immediately when a stop request is initiated
• onStoppingEnd - Called once the current response stops playing

Getting Background Video

Retrieve the idle background video URL for use in welcome screens or widget mode:

const videoUrl = await Conversation.getBackgroundVideo({
  orgId: "your-org-id",
  headId: "your-head-id",
});

Instance Methods

startSession()

Start the conversation session and begin audio & video playback:

await conversation.startSession();

This method should be called after user interaction to ensure audio context is properly initialized, especially on mobile browsers.

sendMessage(message)

Send a text message to the digital human:

conversation.sendMessage("Hello, how are you?");

toggleMicrophone()

Toggles microphone status between ON/OFF.

conversation.toggleMicrophone();

getMicrophoneStatus()

Get status of microphone. Can be ON/OFF/PROCESSING

conversation.getMicrophoneStatus();

keepSession()

Sends keep-alive event to prevent session timeout. Trigger this when you receive the onTimeoutWarning event to prevent session from timing out.

conversation.keepSession();

This clears both audio and video queues and returns the digital human to idle state.

toggleMute()

Toggle the mute status of the audio output:

const volume = await conversation.toggleMute();
console.log("New volume:", volume); // 0 for muted, 1 for unmuted

getUserId()

Get the current user's ID:

const userId = conversation.getUserId();

stopResponse()

Stops the ongoing response and notifies you via two callbacks:

• onStoppingStart() - Called immediately when stop is initiated.
• onStoppingEnd() - Called once the current response is stopped.

await conversation.stopResponse();

endSession()

End the conversation session and clean up resources:

await conversation.endSession();

This closes the WebSocket connection, releases the wake lock, and destroys audio/video outputs.

Error Handling

Always handle errors appropriately:

try {
  const conversation = await Conversation.startDigitalHuman({
    orgId: "your-org-id",
    headId: "your-head-id",
    element: videoElement,
    onError: ({ message, endConversation, type }) => {
      if (type === "toast") {
        // Show toast notification
        showToast(message);
        if (endConversation) {
          // Restart the session
          restartSession();
        }
      } else if (type === "modal") {
        // Show modal dialog
        showModal(message);
      }
    },
  });
} catch (error) {
  console.error("Failed to start digital human:", error);
}

TypeScript Support

Full TypeScript types are included:

Development

Please refer to the README.md file in the root of this repository.