Control the Agent

Once your app is connected to the LiveKit room (see Quickstart), all commands to the agent are sent via sendText() on a data-channel topic. This guide covers the three command topics and their payload formats.

Note: If you are using the D-ID Agents SDK or Embed Code, these details are handled internally — call speak() and chat() instead.

Commands

Command	Topic	Payload
`speak`	`did.speak`	`{ "script": { "type": "text", "input": "..." } }`
`chat`	`lk.chat`	plain text string
`interrupt`	`did.interrupt`	empty string

speak

Generates video from text (equivalent to the V1 Talks/Clips Streams API).

Topic: did.speak
Payload: { "script": { "type": "text", "input": "Your text here" } }
Use for: scripted content, announcements, predetermined messages, faster generation without LLM processing

chat

Sends a conversational message. The agent's LLM processes the input and responds.

Topic: lk.chat
Payload: plain text string
Use for: natural conversations, Q&A, context-aware responses using a knowledge base

interrupt

Interrupts the current video sequence and returns the agent to idle.

Topic: did.interrupt
Payload: empty string
Use for: stopping a long response, barge-in flows. Requires interrupt_enabled on the agent (default true).

Sending commands

   // speak — generate video from text
   await room.localParticipant.sendText(
     JSON.stringify({ script: { type: "text", input: "Hello!" } }),
     { topic: "did.speak" }
   );

   // chat — conversational message
   await room.localParticipant.sendText(
     "What is the weather?",
     { topic: "lk.chat" }
   );

   // interrupt — stop the current video
   await room.localParticipant.sendText("", { topic: "did.interrupt" });