---
title: "How to Test and Configure AI Agents in the Playground | Hostman Docs"
description: "Use the Playground to evaluate AI agent performance, improve prompts, and fine-tune settings like temperature, token limits, Top P, search refinement, and penalty parameters to achieve the desired behavior."
---

> For the complete documentation index for AI agents, see [llms.txt](https://hostman.com/llms.txt).

The **Playground** is a testing environment where you can check how your AI agent works and fine-tune it according to your requirements.

## Test the Agent

To test an agent:

1.  Go to the **AI Services** → **Agents** section and select the agent.
2.  Open the **Playground** tab. You will see the prompt defined when the agent was created, the agent settings, and below that a chat interface for interacting with the agent.
3.  Start a conversation with the agent to evaluate the accuracy and speed of its responses, as well as the level of detail and communication style.
4.  If you need to adjust the agent’s behavior, modify its parameters.

![Ambitious Aquila 2026 08 19 12 31](https://content.hostman.com/assets/00d3d480-8b07-43fe-b264-a0a908400033.png?width=1588&height=1790)

## Modify Prompt

Refine the wording in the system prompt or add any missing requirements. Our [prompt guidelines](https://hostman.com/docs/ai-agents/manage-agents/prompts/#prompt-guidelines) can help you with this.

Click **Save changes** to apply the new prompt, then test the agent again.

## Modify Agent Settings

In the panel on the right, you can configure the agent parameters:

-   **Maximum tokens**. Specifies the maximum number of tokens the model can use in a response. Higher values allow the model to generate longer responses.
-   **Temperature**. Controls the creativity of the model. Higher values result in more creative and unpredictable (and sometimes less accurate) responses.
-   **Top P**. Defines the probability threshold for token selection. Higher values produce more diverse responses, while lower values make responses more focused and coherent.
-   **Refine search query**. If enabled, the agent reformulates your request to perform a more accurate search in the knowledge base. Disable this option to search strictly by the exact words and phrases entered.
-   **Presence penalty**. Reduces repetition by penalizing tokens that have already appeared in the text. Higher values result in fewer repetitions.
-   **Frequency penalty**. Reduces repetition based on how frequently tokens appear in the text. Higher values result in fewer repetitions.
-   **Thinking mode**. Enables the reasoning mode for the model

Click **Save changes** to apply the selected parameters to the agent, then test its performance.