Hard time debugging complex AgentSession flows (tools, handoffs, transcripts)

Hi everyone,

While building more complex agents with the LiveKit Agents SDK, I’ve been finding it difficult to understand what’s happening inside an AgentSession while it’s running.

Once agents start doing things like:

  • multiple tool calls

  • agent handoffs

  • streaming transcripts

  • state transitions (listening → thinking → speaking)

debugging becomes pretty hard. I usually end up relying on console logs or digging through events after the run, which gets messy quickly.

LiveKit’s Agent Observability is great for post-run insights and metrics, but during development I often wish there was something like a live debugger for the agent session.

So I built Liveflow.

It’s a VS Code extension + Python package that visualizes your LiveKit agent in real time when it runs.

No code changes required.

It intercepts AgentSession events and shows things like:

  • agent handoffs

  • tool execution timeline

  • live transcripts

  • LLM context

  • agent state

What it shows

  • Agent graph - visualize agents, handoffs, and tool calls

  • Tool timeline - every function tool execution

  • Live transcripts - speech → text streaming

  • Chat context inspector - full LLM context window

  • Agent state - listening / thinking / speaking

To get started:
https://liveflow.lakshyaworks.dev/

Or directly install the Liveflow extension in your preferred IDE.