Introduction

Overview

LiveKit provides tools to test agents during development, verify behavior on every change, and observe real sessions in production. This section covers Agent Console for interactive debugging, unit tests and Agent Simulations for automated checks, and agent observability for sessions with real users.

These tools cover different levels of a conversation and give you different levels of control. Turn-level assertions catch specific regressions. Whole-conversation runs evaluated by an LLM judge catch behavior that emerges across multiple turns. Production observability reveals behavior that isn't covered by tests.

The testing lifecycle

Each stage catches a different class of problems, from issues you find while developing to behavior that only appears in production.

To speak to a local agent from your terminal instead of the dashboard, use console mode in the LiveKit CLI.

Observing agents in production

Tests verify how an agent handles the cases you define. Observability shows what happens in real sessions, including cases your tests don't cover. LiveKit Cloud records each session as a transcript, a trace, and a set of metrics. You can inspect this data in the dashboard, collect it in your own systems, or export it to a third-party backend.

Observability focuses on the conversation: what the agent said, whether a tool call succeeded, and how the session performed. For information about the service running the conversation, such as whether a process crashed or whether logs reach your monitoring provider, see the Logs and Log drains topics in Manage & Deploy.

Third-party testing tools

First-party testing covers text and audio conversations against your own agent. For broader evaluation suites, load testing, or production monitoring beyond what LiveKit provides, consider these third-party services:

In this section