The Playground is your testing environment where you can interact with your agent, adjust settings, and perfect responses before going live.

What You Can Do

  • Test conversations with your agent
  • Adjust AI model and settings
  • Test voice capabilities
  • Manage knowledge base sources
  • Fine-tune system prompts

Prerequisites

  • An agent created in your workspace

Playground Interface

The Playground consists of several key areas:
AreaPurpose
Chat WindowTest conversations with your agent
Settings PanelConfigure AI model, temperature, etc.
Knowledge BaseManage training data sources
Voice SettingsConfigure voice capabilities
screenshot: playground overview

Step-by-Step: Test Your Agent

1

Navigate to Playground

Go to your agent’s Playground tab.
2

Start a Conversation

Type a message in the chat input and press Enter.
3

Review Response

Evaluate the agent’s response for accuracy and tone.
4

Iterate

Adjust settings or system prompt and test again.

AI Model Settings

Selecting a Model

Choose the AI model that powers your agent:
ModelBest ForSpeedCost
GPT-4Complex reasoningMediumHigher
GPT-3.5General useFastLower
GroqUltra-fast responsesVery FastMedium
ClaudeLong conversationsMediumMedium
GeminiMultimodal tasksFastMedium

Temperature Setting

Controls response creativity:
RangeBehaviorUse Case
0.0-0.3Factual, consistentSupport, FAQ
0.4-0.7BalancedGeneral use
0.8-1.0Creative, variedCreative writing
Start with 0.5-0.7 for most use cases and adjust based on testing.

Max Tokens

Maximum response length:
  • Short responses: 200-500 tokens
  • Medium responses: 500-1000 tokens
  • Long responses: 1000-2000 tokens

Knowledge Base Management

Access and manage your agent’s training data:
1

Open Knowledge Base

Click the Knowledge Base button in the Playground.
2

View Sources

See all connected knowledge sources.
3

Add Sources

Click Add Source to upload files or crawl websites.
4

Remove Sources

Click the delete icon to remove a source.

Source Types

  • Files: PDF, DOCX, TXT, CSV
  • Websites: Crawl and index web pages
  • FAQ: Structured Q&A pairs
  • Raw Text: Direct text input
screenshot: knowledge base dialog

Voice Testing

Test voice capabilities in the Playground:
1

Enable Voice

Click the Voice button in the Playground.
2

Configure Settings

Select voice provider, language, and style.
3

Start Voice Chat

Click the microphone to begin speaking.
4

Test Responses

Listen to the agent’s voice responses.

Voice Providers

ProviderFeatures
Google TTSNatural voices, many languages
AWS PollyNeural voices, SSML support
ElevenLabsUltra-realistic voices
Microsoft SpeechEnterprise-grade quality

System Prompt Editing

Refine your agent’s behavior:
1

Open Settings

Click the settings icon in the Playground.
2

Edit System Prompt

Modify the system prompt text.
3

Save Changes

Click Save to apply.
4

Test

Send test messages to verify the changes.

System Prompt Tips

Good system prompt structure:

1. Role definition
   "You are a customer support agent for [Company]..."

2. Capabilities
   "You can help with: orders, returns, product questions..."

3. Limitations
   "You cannot: process payments, access personal data..."

4. Tone guidelines
   "Be friendly, professional, and concise..."

5. Special instructions
   "If asked about pricing, direct to the pricing page..."

Best Practices

Try unusual questions, typos, and off-topic messages to see how your agent handles them.
Use actual customer questions from your support history.
Make small changes and test frequently rather than large rewrites.
If deploying to multiple channels, test in each one.
Keep notes on effective system prompt patterns.

Troubleshooting

  • Check that an AI model is selected
  • Verify API keys are configured
  • Check your credit balance
  • Add more specific knowledge base content
  • Refine the system prompt
  • Lower the temperature setting
  • Ensure sources are processed (not pending)
  • Check that content is relevant to the question
  • Try rephrasing the question
  • Check microphone permissions
  • Verify voice provider is configured
  • Test with a different browser