📊 SuperOptiX Observability & Tracing Guide

Welcome to the comprehensive guide for SuperOptiX's observability and monitoring system! This guide combines the complete overview, real-world experiment, and quick reference for monitoring, debugging, and analyzing your AI agents using the super observe command.

🎯 Overview

SuperOptiX provides a powerful observability system that automatically traces agent execution, tool usage, LLM calls, and performance metrics. The observability system helps you:

Monitor agent performance in real-time
Debug issues with detailed trace analysis
Analyze patterns and optimize agent behavior
Track tool usage and LLM interactions
Export trace data for external analysis

🚀 Quick Start

Let's walk through a complete example of setting up observability for an agent:

Step 1: Initialize a Project and Create an Agent

Bash

# Initialize a new SuperOptiX project
super init my_project
cd my_project

# Pull a pre-built agent (Genies tier for full observability features)
super agent pull developer --tier genies

# Compile the agent to generate the pipeline
super agent compile developer

Step 2: Run the Agent to Generate Traces

Bash

# Run the agent - this automatically enables tracing
super agent run developer --goal "Write a simple Python function to calculate the factorial of a number"

The agent execution automatically generates trace files in .superoptix/traces/ with detailed information about: - Model initialization - Tool setup and execution - Pipeline execution flow - Performance metrics - Error handling

Step 3: Explore Observability Features

Bash

# List all agents with available traces
super observe list

# View traces for a specific agent
super observe traces developer_20250714_200501

# Launch the interactive dashboard
super observe dashboard --auto-open

# Analyze performance metrics
super observe analyze developer_20250714_200501 --days 1

📋 Available Commands

`super observe list` - List Agents with Traces

Lists all agents that have generated trace files in your project.

Bash

super observe list

Output Example:

Text Only

📋 Available Agents with Traces

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━┓
┃ Agent ID                  ┃ Trace Count ┃ Last Activity       ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━┩
│ developer                 │ 2           │ 2025-07-14 20:05:11 │
│ developer_20250714_200501 │ 11          │ 2025-07-14 20:05:11 │
└───────────────────────────┴─────────────┴─────────────────────┘

`super observe traces` - View Agent Traces

View detailed execution traces for a specific agent.

Bash

# Basic trace view
super observe traces <agent_id>

# Detailed analysis with tool and LLM information
super observe traces <agent_id> --detailed --show-tools --show-llm

# Filter by component or status
super observe traces <agent_id> --component pipeline --status success

# Export traces to JSON or CSV
super observe traces <agent_id> --export json --output traces.json

Options: - --component: Filter traces by component (pipeline, tool, execution, etc.) - --status: Filter by status (success, error, warning, info) - --limit: Limit number of traces shown (default: 100) - --detailed: Show detailed trace analysis - --show-tools: Show tool execution details - --show-llm: Show LLM call details - --export: Export format (json, csv) - --output: Output file path

Output Example:

Text Only

🔍 Loading traces for agent: developer_20250714_200501
✅ Loaded 11 trace events
              Traces for Agent: developer_20250714_200501               
┏━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━┳━━━━━━━━━━━┓
┃ Time     ┃ Component ┃ Event                   ┃ Status  ┃ Duration  ┃
┡━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━╇━━━━━━━━━━━┩
│ 20:05:01 │ pipeline  │ model_initialized       │ success │ -         │
│ 20:05:01 │ pipeline  │ tools_setup_start       │ success │ -         │
│ 20:05:01 │ pipeline  │ tools_initialized       │ success │ -         │
│ 20:05:01 │ pipeline  │ tools_setup_end         │ success │ 0.5ms     │
│ 20:05:01 │ pipeline  │ react_agent_initialized │ success │ -         │
│ 20:05:01 │ execution │ pipeline_forward_start  │ success │ -         │
│ 20:05:05 │ tool      │ calculate_start         │ success │ -         │
│ 20:05:05 │ calculate │ tool_execution_success  │ success │ -         │
│ 20:05:05 │ tool      │ calculate_end           │ success │ 0.1ms     │
│ 20:05:11 │ pipeline  │ execution_completed     │ success │ -         │
│ 20:05:11 │ execution │ pipeline_forward_end    │ success │ 10270.0ms │
└──────────┴───────────┴─────────────────────────┴─────────┴───────────┘

📊 Total: 11 | Errors: 0 | Avg: 3423.5ms

`super observe dashboard` - Launch Interactive Dashboard

Launch a web-based dashboard for real-time monitoring and analysis.

Bash

# Launch dashboard with default settings
super observe dashboard

# Launch with custom port and auto-open browser
super observe dashboard --port 8502 --host localhost --auto-open

# Monitor a specific agent
super observe dashboard --agent-id developer_20250714_200501

Options: - --port: Dashboard port (default: 8501) - --host: Dashboard host (default: localhost) - --auto-open: Automatically open browser - --agent-id: Monitor specific agent

The dashboard provides: - Real-time trace visualization - Performance metrics and charts - Tool usage analytics - Error tracking and debugging - Memory and context analysis

`super observe analyze` - Performance Analysis

Analyze agent performance over time and generate insights.

Bash

# Analyze last 7 days (default)
super observe analyze <agent_id>

# Analyze specific time period
super observe analyze <agent_id> --days 30

Output Example:

Text Only

📊 Performance Summary for Agent: developer_20250714_200501
 Total Events              11         
 Successful Events         11         
 Error Events              0          
 Warning Events            0          
 Average Duration          3423.5 ms  
 Median Duration           0.5 ms     
 95th Percentile Duration  10270.0 ms

`super observe check` - Trace Configuration Check

Check pipeline tracing configuration and verify trace generation.

Bash

# Check trace configuration
super observe check

# Check with test run
super observe check --agent-id <agent_id> --run-test

# Check DSPy configuration
super observe check --check-dspy

Options: - --agent-id: Test specific agent - --run-test: Run a test agent execution - --check-dspy: Check DSPy configuration

`super observe debug` - Interactive Debugging

Start an interactive debugging session for an agent.

Bash

# Start debug session
super observe debug agent <agent_id>

# Enable step-by-step debugging
super observe debug agent <agent_id> --enable-step-mode

# Break on errors or memory operations
super observe debug agent <agent_id> --break-on-error --break-on-memory

Debug Commands: - help: Show available commands - continue: Continue execution - step: Step through execution - breakpoint <component>: Set breakpoint - inspect: Inspect current state - memory: View memory contents - trace: Show execution trace - export: Export debug data

📊 Trace Data Structure

SuperOptiX generates comprehensive trace data in JSONL format. Each trace event contains:

JSON

{
  "event_id": "unique-event-identifier",
  "timestamp": "2025-07-14T20:05:01.215246",
  "event_type": "model_initialized",
  "component": "pipeline",
  "data": {
    "model": "llama3.1:8b",
    "provider": "ollama",
    "tier": "genies",
    "adapter": "ChatAdapter"
  },
  "parent_id": null,
  "duration_ms": null,
  "status": "success",
  "metadata": null
}

Event Types: - model_initialized: Model setup and configuration - tools_setup_start/end: Tool initialization - tools_initialized: Tool registration - react_agent_initialized: ReAct agent setup - pipeline_forward_start/end: Main execution flow - calculate_start/end: Tool execution - tool_execution_success: Tool results - execution_completed: Pipeline completion

Components: - pipeline: Core pipeline operations - execution: Main execution flow - tool: Tool execution - calculate: Calculator tool - text_analyzer: Text analysis tool - file_reader: File reading tool

🔬 Real-World Experiment: Developer Agent

This section documents our comprehensive experiment with the SuperOptiX observability system using a developer agent.

🎯 Experiment Overview

Goal: Document and demonstrate the super observe command functionality by creating a test agent, running it, and analyzing the generated traces.

Agent Used: Developer Assistant (Genies tier) Task: Write a simple Python function to calculate the factorial of a number

🚀 Experiment Steps

Step 1: Project Initialization

Bash

# Initialize a new SuperOptiX project
super init swe
cd swe

Result: Created a new SuperOptiX project with the .super file marker.

Step 2: Agent Creation

Bash

# Pull a pre-built developer agent with Genies tier
super agent pull developer --tier genies

Result: Successfully added the Developer Assistant agent with: - Name: Developer Assistant - Industry: Software - Tier: Genies - Features: ReAct Agents + Tools + Memory - Location: swe/agents/developer/playbook/developer_playbook.yaml

Step 3: Agent Compilation

Bash

# Compile the agent to generate executable pipeline
super agent compile developer

Result: Generated a Genies-tier pipeline with: - Framework: DSPy (Mixin template) - Features: ReAct, Tools, RAG Support, Memory - Output: swe/agents/developer/pipelines/developer_pipeline.py - BDD Scenarios: 5 found for testing

Step 4: Agent Execution

Bash

# Run the agent with a specific goal
super agent run developer --goal "Write a simple Python function to calculate the factorial of a number"

Result: Agent executed successfully with: - Agent ID: developer_20250714_200501 - Model: llama3.1:8b (Ollama backend) - Tools: 3 tools configured (calculator, text_analyzer, file_reader) - Execution Time: 10.27 seconds - Status: Success ✅

📊 Trace Analysis Results

Trace File Generation

The agent execution automatically generated trace files in .superoptix/traces/:

Text Only

📁 Trace Files Generated:
├── .superoptix/traces/developer_20250714_200501.jsonl (4057 bytes)
└── .superoptix/traces/developer.jsonl (960 bytes)

Key Trace Events Observed

Model Initialization: model_initialized - Model setup with Ollama backend
Tool Setup: tools_setup_start/end - Tool initialization (0.5ms)
Tool Registration: tools_initialized - 3 tools registered
ReAct Setup: react_agent_initialized - ReAct agent configuration
Execution Start: pipeline_forward_start - Main execution begins
Tool Execution: calculate_start/end - Calculator tool used (0.1ms)
Execution Complete: pipeline_forward_end - Total execution (10270ms)

Performance Insights

Total execution time: 10.27 seconds
Tool execution time: 0.1ms (calculator tool)
Setup overhead: 0.5ms for tool initialization
LLM processing: Majority of time spent in LLM inference

🎯 Key Findings

1. Automatic Tracing

Tracing is automatic: No manual configuration required
Comprehensive coverage: Captures model, tools, execution, and performance data
Structured format: JSONL format for easy parsing and analysis

2. Rich Trace Data

Event hierarchy: Parent-child relationships between events
Timing information: Precise duration measurements
Component separation: Clear separation of pipeline, execution, and tool events
Status tracking: Success, error, warning, and info statuses

3. Tool Usage Analysis

Tools used: Calculator tool attempted (with syntax error)
Tool integration: Seamless integration with ReAct framework
Error handling: Graceful handling of tool execution errors

⚡ Quick Reference

🚀 Essential Commands

List Agents with Traces

Bash

super observe list

View Agent Traces

Bash

# Basic trace view
super observe traces <agent_id>

# Detailed analysis
super observe traces <agent_id> --detailed --show-tools --show-llm

# Filter by component
super observe traces <agent_id> --component pipeline

# Export traces
super observe traces <agent_id> --export json --output traces.json

Launch Dashboard

Bash

# Default dashboard
super observe dashboard

# Custom port and auto-open
super observe dashboard --port 8502 --auto-open

# Monitor specific agent
super observe dashboard --agent-id <agent_id>

Performance Analysis

Bash

# Analyze last 7 days (default)
super observe analyze <agent_id>

# Analyze specific period
super observe analyze <agent_id> --days 30

Debug Agent

Bash

# Start debug session
super observe debug agent <agent_id>

# Step-by-step debugging
super observe debug agent <agent_id> --enable-step-mode

# Break on errors
super observe debug agent <agent_id> --break-on-error

Check Configuration

Bash

# Check trace configuration
super observe check

# Test with specific agent
super observe check --agent-id <agent_id> --run-test

📋 Command Options Summary

`super observe traces` Options

--component: Filter by component (pipeline, tool, execution)
--status: Filter by status (success, error, warning, info)
--limit: Limit number of traces (default: 100)
--detailed: Show detailed analysis
--show-tools: Show tool execution details
--show-llm: Show LLM call details
--export: Export format (json, csv)
--output: Output file path

`super observe dashboard` Options

--port: Dashboard port (default: 8501)
--host: Dashboard host (default: localhost)
--auto-open: Automatically open browser
--agent-id: Monitor specific agent

`super observe analyze` Options

--days: Number of days to analyze (default: 7)

`super observe debug` Options

--enable-step-mode: Enable step-by-step debugging
--break-on-error: Break on error
--break-on-memory: Break on memory operations

🔍 Debug Commands

Once in debug mode: - help: Show available commands - continue: Continue execution - step: Step through execution - breakpoint <component>: Set breakpoint - inspect: Inspect current state - memory: View memory contents - trace: Show execution trace - export: Export debug data

🎯 Common Event Types

model_initialized: Model setup
tools_setup_start/end: Tool initialization
tools_initialized: Tool registration
react_agent_initialized: ReAct setup
pipeline_forward_start/end: Main execution
calculate_start/end: Tool execution
tool_execution_success: Tool results
execution_completed: Pipeline completion

🔧 Common Components

pipeline: Core pipeline operations
execution: Main execution flow
tool: Tool execution
calculate: Calculator tool
text_analyzer: Text analysis tool
file_reader: File reading tool

🔧 Advanced Features

External Tracing Integrations

SuperOptiX supports integration with external tracing systems:

MLflow: For experiment tracking and model management
Langfuse: For LLM application monitoring
Custom tracers: Extensible tracing framework

Performance Monitoring

The observability system tracks: - Execution duration and timing - Tool usage patterns - LLM call performance - Error rates and types - Memory usage trends - Component-specific metrics

Debugging Capabilities

Step-by-step execution: Walk through agent execution
Breakpoints: Set breakpoints on specific components
State inspection: Examine agent state at any point
Memory analysis: View memory contents and operations
Error tracking: Detailed error analysis and debugging

🎯 Best Practices

1. Enable Tracing for All Agents

Always run agents with tracing enabled to capture comprehensive execution data:

Bash

# Tracing is automatically enabled for all agent runs
super agent run <agent_id> --goal "your goal"

2. Regular Performance Analysis

Schedule regular performance analysis to identify optimization opportunities:

Bash

# Weekly performance review
super observe analyze <agent_id> --days 7

# Monthly trend analysis
super observe analyze <agent_id> --days 30

3. Monitor Tool Usage

Track tool usage patterns to optimize agent capabilities:

Bash

# View tool execution details
super observe traces <agent_id> --show-tools --detailed

4. Debug Production Issues

Use the debugging system to troubleshoot production problems:

Bash

# Interactive debugging session
super observe debug agent <agent_id> --break-on-error

5. Export Trace Data

Export trace data for external analysis and reporting:

Bash

# Export to JSON for analysis
super observe traces <agent_id> --export json --output analysis.json

# Export to CSV for spreadsheet analysis
super observe traces <agent_id> --export csv --output data.csv

🚨 Troubleshooting

No Traces Found

If you don't see any traces:

Verify agent execution: Ensure the agent has been run at least once
Check project structure: Make sure you're in a SuperOptiX project directory
Verify trace directory: Check .superoptix/traces/ for trace files
Run trace check: Use super observe check to diagnose issues

Dashboard Not Starting

If the dashboard fails to start:

Check port availability: Ensure port 8501 (or your chosen port) is free
Verify Streamlit installation: Install with pip install streamlit
Check permissions: Ensure write access to /tmp/ directory
Review logs: Check for error messages in the console output

Performance Issues

If you experience performance problems:

Limit trace data: Use --limit to reduce trace output
Filter traces: Use --component and --status filters
Export selectively: Export only necessary data
Monitor disk usage: Clean up old trace files periodically

📈 Performance Metrics

Key Metrics to Monitor

Total Events: Number of trace events
Success Rate: Percentage of successful events
Error Rate: Percentage of error events
Average Duration: Mean execution time
95^th Percentile: Performance threshold
Tool Usage: Frequency of tool calls
LLM Calls: Number of model interactions

Typical Values

Setup Time: 0.1-1.0ms
Tool Execution: 0.1-10ms
LLM Processing: 1-30 seconds
Total Execution: 1-60 seconds

🎉 Quick Start Workflow

Run an agent to generate traces:

Bash

super agent run <agent_id> --goal "your task"

List available traces:
Bash
```
super observe list
```

View traces for analysis:

Bash

super observe traces <agent_id> --detailed

Launch dashboard for monitoring:
Bash
```
super observe dashboard --auto-open
```

Analyze performance:

Bash

super observe analyze <agent_id> --days 7

Debug issues if needed:

Bash

super observe debug agent <agent_id> --break-on-error

Agent Development Guide - Learn how to create and customize agents
Tool Development Guide - Build custom tools for your agents
Memory System Guide - Understand agent memory and context management
Evaluation and Testing Guide - Test and validate your agents
CLI Reference - Complete CLI command reference

🎉 Conclusion

The SuperOptiX observability system provides comprehensive monitoring, debugging, and analysis capabilities for your AI agents. By following this guide, you can effectively monitor agent performance, debug issues, and optimize your agentic AI workflows.

Start exploring your agents today with super observe list and discover the power of comprehensive AI observability! 🚀

📊 SuperOptiX Observability & Tracing Guide

🎯 Overview

🚀 Quick Start

Step 1: Initialize a Project and Create an Agent

Step 2: Run the Agent to Generate Traces

Step 3: Explore Observability Features

📋 Available Commands

super observe list - List Agents with Traces

super observe traces - View Agent Traces

super observe dashboard - Launch Interactive Dashboard

super observe analyze - Performance Analysis

super observe check - Trace Configuration Check

super observe debug - Interactive Debugging

📊 Trace Data Structure

🔬 Real-World Experiment: Developer Agent

🎯 Experiment Overview

🚀 Experiment Steps

Step 1: Project Initialization

Step 2: Agent Creation

Step 3: Agent Compilation

Step 4: Agent Execution

📊 Trace Analysis Results

Trace File Generation

Key Trace Events Observed

Performance Insights

🎯 Key Findings

1. Automatic Tracing

2. Rich Trace Data

3. Tool Usage Analysis

⚡ Quick Reference

🚀 Essential Commands

List Agents with Traces

View Agent Traces

Launch Dashboard

Performance Analysis

Debug Agent

Check Configuration

📋 Command Options Summary

super observe traces Options

super observe dashboard Options

super observe analyze Options

super observe debug Options

🔍 Debug Commands

🎯 Common Event Types

🔧 Common Components

🔧 Advanced Features

External Tracing Integrations

Performance Monitoring

Debugging Capabilities

🎯 Best Practices

1. Enable Tracing for All Agents

2. Regular Performance Analysis

3. Monitor Tool Usage

4. Debug Production Issues

5. Export Trace Data

🚨 Troubleshooting

No Traces Found

Dashboard Not Starting

Performance Issues

📈 Performance Metrics

Key Metrics to Monitor

Typical Values

🎉 Quick Start Workflow

📚 Related Documentation

🎉 Conclusion

`super observe list` - List Agents with Traces

`super observe traces` - View Agent Traces

`super observe dashboard` - Launch Interactive Dashboard

`super observe analyze` - Performance Analysis

`super observe check` - Trace Configuration Check

`super observe debug` - Interactive Debugging

`super observe traces` Options

`super observe dashboard` Options

`super observe analyze` Options

`super observe debug` Options