Playground
Test your agents before deploying to visitors.
The Playground is your testing environment for agents. Use it to verify responses, fine-tune instructions, and ensure your agent is ready before deploying to visitors.
What is the Playground?
The Playground lets you have conversations with your agents exactly as your visitors would, but in a safe testing environment. You can:
- Test how agents respond to different questions.
- Verify that knowledge is being used correctly.
- See source citations and reasoning.
- Experiment without affecting live visitors.
Using the Playground
Getting started
- Navigate to Playground in the sidebar.
- Select an agent to test.
- Optionally select a channel (to test channel-specific settings).
- Start typing questions.
Testing conversations
Ask questions your visitors might ask.
| Type | Example questions |
|---|---|
| Basic info | "What are your hours?" |
| Specific details | "Do you have wheelchair access?" |
| Complex questions | "I'm visiting with kids under 5, what do you recommend?" |
| Edge cases | "Can I bring my dog?" |
| Off-topic | "What's the weather like?" |
What to look for
| Check | Why it matters |
|---|---|
| Accuracy | Is the information correct? |
| Relevance | Does the response answer the question? |
| Tone | Does it match the personality you set? |
| Sources | Are the right knowledge paths being used? |
| Length | Is the response too long or too short? |
| Edge cases | How does it handle questions it can't answer? |
Playground features
Source citations
See which Knowledge Paths the agent used:
- Click on a response to see sources.
- Verify the right information was retrieved.
- Identify missing knowledge.
Thinking process
With compatible models (like GreenR), you can see the agent's reasoning:
- How it interpreted the question.
- What information it considered.
- Why it chose a particular response.
Session management
| Action | What it does |
|---|---|
| New Session | Start a fresh conversation. |
| Clear History | Remove current conversation. |
| Switch Agent | Test a different agent. |
| Switch Channel | Test channel-specific settings. |
Testing tips
Start with common questions
Test the questions visitors ask most:
- List your top 10 FAQs.
- Ask each one in the Playground.
- Verify responses are accurate and helpful.
- Adjust knowledge or instructions as needed.
Test edge cases
See how your agent handles tricky situations.
| Scenario | What to test |
|---|---|
| Unknown topics | Questions outside your knowledge base. |
| Ambiguous questions | "How much?" (how much for what?). |
| Multiple questions | "What are your hours and prices?" |
| Follow-up questions | "What about weekends?" (after asking about hours). |
| Complaints | "I'm not happy with my experience". |
Check personality consistency
Make sure the tone stays consistent:
- Ask formal questions.
- Ask casual questions.
- Ask emotional questions.
- The personality should remain consistent.
Test knowledge coverage
Verify all your knowledge is accessible:
- Look at each Knowledge Path you've created.
- Ask a question that should use that knowledge.
- Confirm the agent finds and uses it.
Improving based on testing
If responses are wrong
| Problem | Solution |
|---|---|
| Wrong information | Update the Knowledge Path with correct info. |
| Missing information | Add a new Knowledge Path. |
| Conflicting answers | Consolidate or clarify knowledge. |
| Outdated content | Refresh URL-based knowledge. |
If responses are off-tone
| Problem | Solution |
|---|---|
| Too formal | Change personality to "friendly" or "casual". |
| Too casual | Change personality to "professional" or "formal". |
| Too long | Add instruction: "Keep responses concise". |
| Too short | Add instruction: "Provide detailed explanations". |
If agent can't answer
| Problem | Solution |
|---|---|
| Topic not covered | Add relevant Knowledge Path. |
| Knowledge not assigned | Assign the Knowledge Path to the agent. |
| Question too vague | Agent should ask for clarification (add to instructions). |
| Intentionally unsupported | Add instruction to politely decline. |
Iterating on your agent
The testing cycle
- Test in Playground.
- Identify issues.
- Update knowledge or instructions.
- Test again.
- Repeat until satisfied.
When to deploy
Your agent is ready when:
- Common questions get accurate answers.
- Edge cases are handled gracefully.
- Tone matches your brand.
- Sources are relevant and correct.
- You're confident in the responses.
Common questions
Is Playground testing visible to visitors?
No! Playground conversations are completely separate. Visitors never see your tests.
Can I share Playground conversations?
Currently, conversations can't be directly shared. Take screenshots or copy text if you need to share with your team.
Why does my agent give different answers to the same question?
AI models have some natural variation. If consistency is important, lower the temperature setting in agent configuration.
How do I test channel-specific settings?
Select the channel in the Playground dropdown. This applies that channel's quick start buttons, welcome messages, and other settings.
Can I test multiple agents at once?
Not in the same session. Open multiple browser tabs to compare agents side by side.
Does Playground count toward my usage?
Playground conversations use the same AI models as live conversations, so they do count toward any usage limits. However, there's no separate limit for testing.