Sub-Agents

Use agents as building blocks within other agents. Sub-agents enable modular, reusable AI workflows.

How It Works

A sub-agent is any Fruxon agent invoked as a step in another agent's workflow:

1. Parent agent calls sub-agent with parameters
2. Sub-agent executes its workflow
3. Sub-agent returns output to parent
4. Parent continues with the result

Adding a Sub-Agent

1. In Agent Studio, click + to add a node
2. Select Sub-Agent
3. Choose the agent to invoke
4. Map parameters from your workflow

Access outputs in subsequent steps:

{{sub_agent_step.summary}}
{{sub_agent_step.result}}

Composition Patterns

Sequential

Chain sub-agents one after another:

extract_text → summarize → analyze_sentiment

Parallel

Run multiple sub-agents concurrently:

     ┌─ sentiment_analyzer
input ├─ keyword_extractor  → merge results
     └─ language_detector

Conditional

Route to different sub-agents based on logic:

If type == "technical" → technical_support
Else → general_support

Best Practices

• Single responsibility - Each sub-agent does one thing well
• Clear interfaces - Well-defined parameters and outputs
• Version pinning - Pin parent agents to specific sub-agent versions
• Test independently - Verify each sub-agent before composing

Next Steps

• Agent Studio - Visual workflow builder
• Versioning - Managing agent versions