Skip to main content

🤖 AI Playground

On this page, you can chat with your AI agents. Test their responses, experiment with different prompts, and interact with your agents in real-time to see how they behave and respond to various inputs.

Agents and Providers​

To start testing your agents, you first need to select an agent from the available agents in your company and a provider.

Prompts page

After you select your agent and provider, you will see the chat window, where you can start chatting with your agent.

💬 Chatting with your Agent​

Once an agent and provider are selected, the chat interface becomes active. Here's how to interact with your agent:

  1. Send a Message: Type your message in the input box at the bottom and press Enter.
  2. Agent Responds: The agent will process your input and generate a response. You may see a "AI is thinking..." indicator while this happens.
  3. Real-time Interaction: Continue the conversation to test different scenarios and see how the agent's personality and knowledge hold up over multiple turns.

🧪 Use Cases for the Playground​

The AI Playground is an essential tool for refining your agents. Here are some common use cases:

  • Personality Testing: Check if your agent's dialog style, motivations, and personality traits are reflected in its responses. For example, if you've configured a "cautious" agent, does it ask clarifying questions?
  • Knowledge Verification: Ask questions related to the documents or text in your agent's Knowledge Collections to ensure it can find and use the information correctly.
  • Prompt Experimentation: Quickly test how changes to a prompt affect the agent's behavior without needing to run a full application build.
  • Debugging and Refinement: Identify unexpected or incorrect responses and use the insights to go back and refine the agent's configuration.

✨ Tips for Effective Testing​

Get the most out of the Playground with these tips:

  • Be Specific: Ask clear, targeted questions to get predictable responses. Vague questions can lead to generic answers.
  • Test Edge Cases: Try to push the agent's boundaries with unexpected, complex, or nonsensical inputs to see how it handles errors.
  • Role-Play: Interact with the agent as your target user would. If it's a training assistant, act like a trainee. If it's a customer support bot, act like a customer.
  • Iterate and Compare: Tweak a single personality trait or prompt, and then run the same test conversation to see the direct impact of your change.

💾 Chat History​

By default, each chat in the AI Playground is a one-time session and is not saved. However, you can configure your environment to save chat history.

  • Session-based: Without configuration, the history is cleared when you refresh the page.
  • Saving History: You can enable chat history persistence, which currently supports saving up to 100,000 characters of conversation. This is useful for longer-term analysis and debugging of agent interactions.