Skip to main content

​
Why Simulate Conversations?

Testing AI conversations manually is time-consuming and often misses edge cases. Maxim supports both agent simulation and prompt simulation to help you:
  • Test how your AI maintains context across multiple exchanges
  • Evaluate responses to different user emotions and behaviors
  • Verify proper use of business context and policies
  • Identify potential conversation dead-ends

​
Agent Simulation

Test your complete agent workflows with AI-generated user interactions. This is particularly valuable for multi-agent systems where complex user scenarios can reveal workflow issues. Agent Simulation

​
Prompt Simulation

Test individual prompts or prompt chains with AI-generated inputs, allowing you to optimize prompt performance before integrating into your agent. Prompt Simulation

​
1. Create a Realistic Scenario and Be Specific About the Situation You Want to Test

  • Customer requesting refund for a defective laptop
  • New user needs help configuring account security settings
  • Customer confused about unexpected charges on their bill

​
Agent Description (for Agent Simulation)

When setting up agent simulation, provide a detailed description of your agent’s purpose, capabilities, and business context. This helps the simulation generate more realistic and relevant test scenarios. Example: 
Financial Advisor Agent

This agent helps users with investment decisions and financial planning. It can:
- Analyze portfolio performance
- Recommend investment strategies based on risk tolerance
- Provide market insights and trends
- Help with retirement planning

The agent maintains a professional, helpful tone and always recommends consulting with a licensed financial advisor for major decisions.
Agent Description

​
2. Define the User Persona

  • Frustrated customer seeking refund
  • New user needing security help
  • Confused customer with billing issues
Mix different emotional states and expertise levels to test how your agent adapts its communication style.
After defining the user persona, select the field where your agent’s replies come from: Response configuration

​
3. Advanced settings (optional)

  • Maximum number of turns: Set a limit for conversation turns. If no value’s set, the simulation ends when complete
  • Reference tools: Attach any tools you want to test with the simulation. You can learn more about setting up tools here
  • Reference context: Add context sources to enhance conversations. Learn more here
Workflow simulation - advanced settings

​
Example Simulation

Here’s a real-world example of a simulated conversation: Live simulation