The Playground is your testing and fine-tuning environment for agent teams. It's where you experiment with agent configurations, test how agents work together, and see exactly what's happening behind the scenes before deploying to production workflows.

What is the Playground?

The Playground is a safe, isolated environment where you can:

Test agent configurations without affecting production agents
See agent reasoning - understand the thought process behind responses
Observe knowledge and memory effects - see how your knowledge base and memory influence responses
Test agent teams - see how multiple agents work together
Fine-tune orchestration patterns - perfect manager-worker relationships before deployment

Think of it as a sandbox where you can experiment freely, then export your perfected configurations directly to production workflows.

Getting Started

Creating a Playground Session

Navigate to the Playground from the main navigation
Click "Create New Session"
Choose your mode:
- Agent Testing Mode: Test individual agents or switch between multiple agents
- Orchestrator Mode: Test manager-worker agent patterns (Pro users only)
Give your session a name (e.g., "Customer Support Team Test")
Click "Create Session"

Session Management

Save sessions: All sessions are automatically saved
Load previous sessions: Access your session history from the sidebar
Delete sessions: Remove sessions you no longer need
Session persistence: Your conversations, configurations, and context are preserved

Agent Testing Mode

Agent Testing Mode lets you test individual agents or switch between multiple agents in the same conversation. This is perfect for:

Comparing how different agents handle the same input
Testing agent configurations before saving
Simulating workflow behavior with context handoffs
Understanding how knowledge and memory affect responses

Testing Individual Agents

Select an agent from the dropdown
Configure the agent (optional):
- Select specific tools to enable
- Toggle web search or image generation
- Adjust other settings as needed
Start chatting to see how the agent responds
Observe the response with full visibility into:
- Agent reasoning (if enabled)
- Tool calls and executions
- Knowledge base citations
- Memory usage indicators

Switching Agents Mid-Conversation

One of the most powerful features is switching agents during a conversation:

Continue your conversation with Agent A
Click "Switch Agent" and select Agent B
Agent B receives the full conversation context
See how Agent B handles the same situation differently

This lets you:

Compare agent performance side-by-side
Test which agent is best for specific tasks
Simulate sequential workflow behavior

Context Handoffs

When switching agents, you can control how context is passed:

With context: The new agent sees the full conversation history
Without context: The new agent starts fresh

This simulates how agents work in sequential workflows, where context can be shared or isolated between steps.

Viewing Agent Reasoning

When agents provide responses, you can see their reasoning process:

Look for the "Agent Reasoning" section below the response
Click to expand and see the agent's thought process
Understand why the agent made specific decisions
Use this to refine prompts and configurations

[!TIP] Understanding Reasoning Agent reasoning shows the step-by-step thought process. This is invaluable for debugging why an agent made a particular decision and for improving your prompts.

Observing Knowledge and Memory Effects

The Playground shows you exactly how knowledge and memory affect agent responses:

Knowledge Base Effects:

Citations: When the agent references your knowledge base, you'll see source citations
Relevant documents: See which documents were retrieved and used
Website content: If you've indexed websites, see when that content is referenced

Memory Effects:

Memory indicators: Visual indicators show when the agent is using stored memories
Context from past conversations: See how previous interactions influence current responses
Memory extraction: Watch as the agent identifies and stores new information

This visibility helps you:

Understand if your knowledge base is being used effectively
See which memories are being recalled
Optimize your knowledge base and memory settings

Tool Execution Visibility

When agents use tools, you see everything:

Tool selection: Which tool the agent chose to use
Tool execution: Real-time execution of tool calls
Tool results: The data returned from tools
Tool approval: If manual approval is enabled, approve or reject tool calls

This lets you:

Debug tool integrations
Understand when and why tools are used
Test tool combinations before production

Orchestrator Mode (Pro Users)

Orchestrator Mode is available to Pro users and provides deep insights into manager-worker agent patterns. This mode lets you:

Configure a manager agent with delegation logic
Add multiple worker agents to a team
See exactly how the manager makes decisions
Watch each agent execute step-by-step
Understand the full orchestration flow

Setting Up Orchestrator Mode

Create a new session and select "Orchestrator Mode"
Configure the Manager Agent:
- Select which agent will act as the manager
- Set the manager's system prompt and instructions
- Configure delegation logic and decision-making rules
Add Worker Agents:
- Select agents from your available agents
- Configure each worker agent's tools and settings
- Build your agent team
Set Execution Parameters:
- Maximum iterations
- Timeout settings
- Other orchestration controls

Understanding Manager Decisions

In Orchestrator Mode, you see exactly how the manager works:

Manager Analysis:

The manager receives your goal or instruction
It analyzes the available worker agents
It decides which agent should handle the task
You see the manager's reasoning for each decision

Delegation Process:

The manager selects a worker agent
It provides specific instructions to that agent
The worker agent executes the task
Results are returned to the manager
The manager evaluates and decides the next step

Execution Modes

Orchestrator Mode offers two ways to observe execution:

Run Mode:

Execute the entire orchestration from start to finish
See the final result after all agents have completed their work
Best for testing complete workflows
Faster execution

Step-by-Step Mode:

Watch each agent execute one at a time
See manager decisions individually
Observe worker outputs as they happen
Best for debugging and understanding the flow
More detailed visibility

What You See in Orchestrator Mode

Manager View:

Manager's analysis of the goal
Available agents and their capabilities
Decision-making process
Instructions given to worker agents

Worker Agent View:

Which agent was selected
Instructions received from manager
Agent's execution and reasoning
Results returned to manager

Overall Flow:

Complete orchestration timeline
Iteration count and progress
Context passed between agents
Final result and summary

This level of visibility helps you:

Debug orchestration logic
Optimize manager prompts
Understand agent interactions
Perfect delegation patterns

Configuration Options

In the Playground, you can configure everything about your agents:

Agent Selection

Choose from all your available agents
Switch agents instantly during testing
Test different agent combinations

Tool Configuration

Enable or disable specific tools
Test tool combinations
See tool execution in real-time
Configure tool approval modes

Knowledge Base

Attach knowledge bases to agents
See when knowledge is retrieved
Observe citation patterns
Test different knowledge configurations

Memory Settings

Enable or disable memory
See memory extraction in action
Observe memory recall
Test memory configurations

Model Selection

Switch between AI models
Compare model performance
Test model-specific features

System Prompts

Modify prompts in real-time
Test prompt variations
See immediate effects of changes

Exporting from Playground

Once you've perfected your agent configuration in the Playground, you can export it:

Export to Workflow

Convert your playground session directly into a production workflow:

Click "Export to Workflow"
Configure workflow settings (triggers, schedule, etc.)
Your agent configurations are preserved
Deploy to production immediately

Save as Template

Create a reusable template from your configuration:

Click "Save as Template"
Name your template
Use it to quickly create new agents or workflows
Share templates with your team

Save as Draft

Keep your work for later refinement:

Your session is automatically saved
Return anytime to continue testing
Refine configurations over time
Export when ready

Best Practices

Start Simple

Begin with basic configurations and gradually add complexity. Test one feature at a time to understand its impact.

Use Reasoning to Improve

Pay attention to agent reasoning. It reveals how agents think and helps you write better prompts.

Test Knowledge and Memory

Actively test how knowledge bases and memory affect responses. This helps you optimize these features.

Compare Agents

Use agent switching to compare different configurations. This helps you find the best setup for each task.

Iterate in Playground

Don't rush to production. Use the Playground to thoroughly test and refine before deploying.

Document Your Findings

Take notes on what works and what doesn't. This helps you build better agent teams over time.

Use Cases

Testing New Agent Configurations

Create a new agent, test it in the Playground, refine based on results, then deploy when satisfied.

Validating Workflow Logic

Simulate entire workflows with context handoffs to ensure agents work together correctly before building the actual workflow.

Debugging Agent Behavior

Use reasoning visibility and step-by-step execution to understand why agents behave in certain ways and fix issues.

Optimizing Knowledge Bases

Test different knowledge base configurations to see which documents and websites provide the best results.

Perfecting Orchestration

Use Orchestrator Mode to fine-tune manager-worker patterns, ensuring optimal delegation and task distribution.

Next Steps

Now that you understand the Playground, you might want to:

The Playground is your gateway to building perfect agent teams. Experiment freely, observe everything, and deploy with confidence!

Using the Playground