Voice AI quickstart

Status: implemented as a Go example, not as a cloud wizard.

This tutorial gets the checked-in voice agent running. It is the fastest source-backed way to verify your LiveKit credentials, model configuration, worker process, agent session, and tools.

Prerequisites

You need:

• Go installed for this repository.
• A LiveKit server or LiveKit Cloud project.
• LIVEKIT_URL, LIVEKIT_API_KEY, and LIVEKIT_API_SECRET.

The example defaults to LiveKit Inference model provider names in basicagent.ConfigFromEnv(). Change those defaults in code, or override the app configuration with the RTP_AGENT_* variables supported by app.DefaultConfigFromEnv().

Run the agent

Set the required LiveKit credentials:

export LIVEKIT_URL=wss://your-project.livekit.cloud
export LIVEKIT_API_KEY=your_api_key
export LIVEKIT_API_SECRET=your_api_secret

Then run the basic agent:

go run ./examples/voice_agents/basic_agent

The program creates the app with basicagent.NewApp(basicagent.ConfigFromEnv()), defers Close, and hands the app's worker server to cli.RunApp.

What the example demonstrates

• app.DefaultConfigFromEnv() reads runtime configuration.
• app.Init() creates the default app, worker server, agent session, and model components.
• agent.NewAgent() creates the custom "Kelly" agent.
• AgentSession.UpdateAgent() installs the custom agent in the running session.
• llm.Tool implementations expose callable tools to the LLM path.
• Agent.OnEnter() can trigger an initial reply with GenerateReplyWithOptions.

Use this example as a starting point for your own app. Copy the structure, not the exact provider defaults, unless the LiveKit Inference models used by the example match your deployment.

Evidence:

• examples/voice_agents/basic_agent/main.go
• examples/voice_agents/basic_agent/basicagent/basic_agent.go
• app/app.go