Overview
Building a voice agent involves more than writing the agent logic. You need to interact with it while you develop, investigate unexpected behavior, and verify that changes don't introduce regressions. The tools in this section support these tasks without requiring a live user session.
Agent Console provides an interactive way to talk to a running agent and inspect its behavior. The CLI debugger lets you reproduce conversations in text mode from a script or coding agent. Unit tests let you turn specific behaviors into repeatable tests that run locally or in CI.
The tools complement each other. Use Agent Console and the CLI debugger to explore agent behavior, then add the cases worth testing to your unit tests so they run with your test suite.
Together, these tools cover the development and testing loop, from exploring an agent during development to verifying behavior as the code changes. For more realistic, multi-turn evaluation with a simulated user, see Agent Simulations. To investigate behavior from sessions with real users, see agent observability.
Choosing a tool
All three tools run without a LiveKit room. They differ in how the conversation is driven and where the agent runs.
| Tool | Who drives the conversation | Where it runs |
|---|---|---|
| Agent Console | You, by voice or by typing. | The browser, against a local or deployed agent. |
| CLI Debugger | A script or coding agent, one command at a time. | Your machine, in text mode. |
| Unit tests | Scripted input and assertions you define. | Your test suite, locally or in CI. |
What to test
Test your agent in the following areas:
- Expected behavior: Does your agent respond with the expected intent and tone for common use cases?
- Tool usage: Are tools called with the expected arguments and context?
- Error handling: How does your agent respond to invalid input or tool failures?
- Grounding: Does your agent stay factual and avoid hallucinating information?
- Misuse resistance: How does your agent respond to attempts to misuse or manipulate it?
Use unit tests for turn-level behaviors such as tool usage and error handling. Use Agent Simulations for behaviors that span multiple turns, such as conversation flow, memory, and misuse resistance.
Unit tests run in text mode, using an LLM through LiveKit Inference or a model plugin. Simulations communicate with your agent using text by default, and can also run over audio to exercise the full STT-LLM-TTS pipeline.
Text mode is a cost-efficient and deterministic way to test agent behavior. Use it while you iterate, and reserve audio runs for turn-taking and speech-specific issues.