Skip to main content
Build a LiveKit agent. Start it. Run chert dev to connect it to FaceTime. You can make a study tutor, interview coach, language partner, or another voice agent using your own code and model provider, with audio or video. Install with npm install -g @trychert/facetime@0.1.0-beta.21 for the version used in this guide. Browser sign-in, automatic LiveKit credential loading from your agent folder, room preparation, and diagnostics are available. Separate chert setup is optional; existing saved settings still take priority. Both chert dev and chert byor support inbound FaceTime Audio and FaceTime Video without a webhook. Use a video-enabled line and an agent configured for the media you need. The voice starter is audio-only. For a video agent, use chert new my-video-agent --video and the video guide. Showing an avatar and seeing the caller’s camera are separate agent capabilities.

Build your first agent

Create a LiveKit account and a complete JavaScript voice agent.

Connect and make a call

Install Chert, sign in, run two terminals, and test your agent.

Audio, video, and caller vision

Publish agent video and connect the caller’s camera to your vision pipeline.

Projects and lines

Understand access, line assignment, and optional saved settings.

Run a hackathon

Prepare accounts, allocate lines, and organize team handoffs.

How the pieces fit together

Your laptop does not become the FaceTime line. You install the CLI on the computer running the connector; installing this npm package on every line computer is not part of this workflow. The CLI does not start your agent code.

Project names and agent names are different

Agent servers, rooms, and named dispatch are LiveKit concepts. Chert uses LiveKit’s dispatch API to connect a call to your code. In our starter, AGENT_NAME=team-otter-agent is read by the code and passed to LiveKit’s ServerOptions.agentName. Starting the program with npm start registers that name with LiveKit. You do not separately create a permanent agent entry in the dashboard for this local workflow. Then run:
Here --agent is the option and team-otter-agent is its value. That value must exactly match the dispatch name registered by your running agent in the same LiveKit project. AGENT_NAME is a convention used by this starter, not an automatic LiveKit setting for every program. See LiveKit’s named dispatch documentation.

What happens during a call

  1. Your agent starts and registers its dispatch name with LiveKit.
  2. chert dev claims the selected Chert line and prepares a waiting room with your agent.
  3. Someone calls that line’s FaceTime address. Chert uses the prepared room for the call.
  4. The caller and agent exchange audio and, on video calls, their configured video tracks through LiveKit and FaceTime.
  5. After hangup, the CLI removes the call’s room and prepares a fresh one.
  6. Ctrl+C stops the connector and releases the line.
A waiting agent proves it joined the room. It does not prove model access or working call audio or video. Test conversation, each required video direction, and cleanup.

Advanced integrations

Most LiveKit developers should use chert dev. If your application must own session creation and cleanup, see custom BYOR callbacks. The separate create-room adapter requires a specific health/cancellation hook. Those integrations have their own manual enrollment and recovery workflow; students do not need them for this guide.