The Agent Builder

Agent nodes are configured directly in your workflow canvas. Double-click any agent node to open its configuration panel.

Creating an Agent Node

1. Click Add Node in your canvas
2. Select Agent from the node type menu
3. Double-click the node to configure it

Agent Configuration

1. Model Selection

Pick your LLM:

• Claude (Anthropic) - Excellent reasoning
• GPT-4 (OpenAI) - Strong general-purpose
• Groq - High-speed inference

Configure parameters: Temperature (creativity level), max tokens (response length).

2. Prompt Configuration

Option A: Use a pre-built template (fastest)

• Browse template library
• Select and apply
• Customize if needed

Option B: Write custom prompt

• Define agent's role and instructions
• Use dynamic variables for flexible content
• Use FlowGenX Xpression to inject workflow data

Example: You are a {agent_role}. Process the customer request: {workflow.customer_message}

3. Memory Configuration

Choose what your agent remembers:

Memory Type:

• Conversation memory - Recent messages in current chat
• Task memory - Progress during multi-step processes
• User memory - Patterns and preferences over time

Memory Scope:

• Execution-level - Agent only remembers its own interactions
• Workflow-level - Agent has full workflow context

4. Input/Output Schema

Define data flow:

• Input Schema - What data structure agent receives
• Output Schema - What data structure agent sends to next node

See the Input/Output Schema page for detailed configuration.

Testing Your Agent

Once configured, don't test in the builder. Instead:

1. Complete your full workflow
2. Click Test to open the Agent Playground
3. Run test inputs and observe behavior
4. Debug any issues
5. Return to canvas to adjust if needed

Best Practices

• Start with React agents - Most versatile for most use cases
• Use templates - Pre-built prompts save time
• Test thoroughly - Use the Playground before production
• Only include necessary tools - More tools = more options for agent to choose from
• Clear prompts - Specific instructions produce better results
• Validate tool connections - Ensure tools work before relying on them

Next Steps