Skip to content

Agent Explorer

The Agent Explorer is a tool designed to run an agent over and over and see what paths it takes in its decision-making process. This is particularly useful for testing and debugging agents, as it allows developers to visualize the agent's behavior in various scenarios.

Getting Started

To use the Agent Explorer, follow these steps:

  1. Install - Go to extend and install the Agent Explorer module.
  2. List - Navigate to the agent list page to see all available agents.
  3. Test - The agent should now have a Explore button, where you can run the agent in the explorer.

How to run a test

  1. Choose an agent from the list.
  2. Write a prompt and add files that you might want to use in the test.
  3. Choose the model you want to use for the test.
  4. Click the Run Agent button to start the test.

Reading the results

On the right side of the screen, you will see the results of the agent's execution. The results are displayed in a table format for each loop iteration. Each row represents a step in the agent's decision-making process, showing the input, output, tools used, and any additional information relevant to that step.

You can click on each row to open it in a new tab it and see more details about the agent's actions, including the raw provider data and the tokenized system prompt.