Skip to content

AI voice agents

An agent is an AI participant. It joins a room, listens to one person and answers out loud. Everyone in the room hears it and can see its state and a live transcript. Use agents for support lines, tutoring, interview practice or a receptionist. Agents need prepaid credit (see Pricing).

Start an agent

bash

curl -X POST https://api.appsocom.dev/rooms/support-42/agents \
  -H "Authorization: Bearer $APPSOCOM_API_KEY" -H "Content-Type: application/json" \
  -d '{"name":"Ava","instructions":"You are Acme'"'"'s support assistant. Keep answers short.","greeting":"Hi, I'"'"'m Ava. How can I help?"}'
# → {"roomId":"support-42","agentId":"agent-…","participant":{…}}

Field
`name`	the name people see
`instructions`	what the agent should do and how to behave (a system prompt)
`greeting`	what it says when it joins
`listenTo`	the `userId` it listens to. By default, the first person with a mic on.
`voice`	which voice to use

All fields are optional. Stop the agent with DELETE /rooms/:roomId/agents/:agentId. It also leaves by itself 20 seconds after the last person does.

In your app

ts

room.agents;                                   // AI participants (participant.kind === "agent")
room.on("agent-state", (agent, state) => …);   // "initializing" | "listening" | "thinking" | "speaking"
room.on("transcript", (seg) => …);             // { id, role: "user" | "agent", speakerId, text, final }
room.sendToAgent(agent.id, "What's my order status?"); // type instead of talking

The agent's voice plays like anyone else's. Transcript segments are sent again with the same id as more words arrive, so replace the line rather than adding a new one.