Prefer a guided walkthrough? ↗ Take the interactive tour
Overview
The Sandbox allows you to safely test your AI agent before exposing it to real users. It simulates real conversations without affecting live channels, customers, or production data. This is the recommended environment for validating behavior, tone, accuracy, and integrations.
What Is the Sandbox?
The Sandbox is a private testing environment where you can:
- Interact with your AI agent as a user
- Test knowledge base answers
- Validate flows, automations, and prompts
- Catch errors before going live
Nothing tested in the Sandbox is visible to end users.
How to Access the Sandbox
- Log in to your ChatAgent dashboard
- Navigate to Agent
- Open the Sandbox environment
You will see a chat interface that mirrors your live agent.
Guided Walkthrough: Testing Your AI Agent in the Sandbox
To make testing easier, we’ve included an interactive guided walkthrough. This guide walks you step by step through using the Sandbox to test your AI agent, from opening the environment to validating responses and edge cases.
Use this walkthrough to:
- Learn where to find the Sandbox
- Understand what to test during conversations
- Follow best practices while interacting with the agent
- Avoid common mistakes before going live
How to Test Your AI Agent
Use the Sandbox to test the following:
1. Knowledge Accuracy
- Ask common customer questions
- Verify answers match your knowledge base
- Check for outdated or missing information
2. Conversation Flow
- Test greetings and first responses
- Try follow-up questions and edge cases
- Ensure the agent keeps context correctly
3. Tone and Brand Voice
- Confirm the agent sounds on-brand
- Adjust prompts if responses feel too robotic or too casual
4. Error Handling
- Ask unclear or incomplete questions
- Verify the agent asks clarifying questions instead of hallucinating
5. Integrations (If Enabled)
- Test actions like ticket creation, handoff, or data lookup
- Confirm the agent responds correctly when integrations fail
Best Practices
- Test with real-world questions from past conversations
- Try both short and long queries
- Re-test after every knowledge or prompt update
- Have multiple team members test independently
When to Go Live
You’re ready to publish when:
- Responses are accurate and consistent
- No critical errors appear in testing
- The tone aligns with your brand
- Edge cases are handled gracefully
Once confirmed, you can confidently enable the agent on live channels.
Need Help?
If you encounter unexpected behavior in the Sandbox, review your knowledge base, prompts, and settings. If issues persist, contact support with screenshots or example queries.