If you want to see the LLM output right in your terminal for each turn, here are a couple of easy ways to do it: • Using Verbose Mode: Just set the

LIVEKIT_EVALS_VERBOSE

environment variable to

1

when you run your tests with pytest. This will give you detailed output for every agent execution, showing both what you typed and the assistant’s response. Here’s how you can do it:

LIVEKIT_EVALS_VERBOSE=1 pytest -s -o log_cli=true <your-test-file>

• Using a Chat Script: Another handy way is to use a bash script that talks to the LLM and shows each turn right in the terminal. For example, the

chat-llama2.sh

script does exactly this by using

curl

to send your questions and then prints both your input and the responses. Here’s a quick snippet:

#!/bin/bash

API_URL=<YOUR_API_URL>

while true; do
    echo -en "\033[0;32m"  # Green color
    read -r -e -p "> " QUESTION
    echo -en "\033[0m"  # Reset color
    # Send the question to the API and print the response
done

Using either of these methods will make it super easy to keep an eye on the conversation flow and debug things right from your terminal. If you run into any hiccups or want to chat more about it, just let me know! Sources: Testing and evaluation | LiveKit Docs | tools/llama-bench/llama-bench.cpp