Advanced Setup

Node-by-node monitoring for specialized use cases. This guide covers the advanced approach for granular control over individual operations.

Recommended: Use Manual Setup with the @tracing decorator for most use cases. This advanced approach is only for very specific scenarios.

When to use advanced setup

Use this only when:

You need to monitor individual LLM calls separately from agent workflows
You’re building custom monitoring tools that require granular operation data
You have complex agent architectures that don’t fit the standard pattern
You need backward compatibility with existing implementations

For most use cases, the simplified agent-level monitoring is recommended.

What you’re doing

You’re manually tracking each operation in your agent workflow, including individual LLM calls and tool executions.

Why this is complex: You must manually track every operation, manage execution IDs, and handle session lifecycle. The @tracing decorator does all this automatically.

Migration to simplified approach

Before (Advanced):


# Complex manual tracking
tracing_response = tracker.start_tracing(agent_name="agent")
execution_id = tracing_response.get("executionId")
try:
    result = await my_function()
    tracker.track_node(input=input_data, output=result, ...)
finally:
    tracker.end_tracing(execution_id=execution_id, agent_name="agent")

After (Simplified):


# Simple decorator
@tracing(agent="agent")
async def my_agent_function():
    result = await my_function()
    return result

Benefits:

90% less code
Automatic operation capture
No manual session management

Next steps

Set up evaluation: Add Quality Assessment so your engineer can detect quality issues.

Enable fixes: Connect GitHub Integration so your engineer can create pull requests with improvements.