Performance Monitor

Summary

The Performance Monitor is a singleton system that tracks and analyzes LLM call performance metrics including latency, success rates, token usage, costs, and error patterns. It provides comprehensive performance analytics for the mindX system.

Technical Explanation

The Performance Monitor implements:

Singleton Pattern: Single instance across the system

Metric Tracking: Per-model, per-task, per-agent metrics

Historical Data: Latency windows and error type tracking

Persistence: JSON-based metric storage

Periodic Saving: Configurable auto-save intervals

Architecture

Type: performance_monitor

Pattern: Singleton

Storage: JSON file-based persistence

Metrics: Comprehensive LLM call tracking

Core Capabilities

LLM call logging and tracking

Latency measurement and analysis

Success/failure rate tracking

Token usage tracking (prompt + completion)

Cost calculation and tracking

Error type classification

Historical metric storage

Periodic metric persistence

Metrics Tracked

Total calls, successful calls, failed calls

Total latency and latency history (deque)

Error types and frequencies

Prompt tokens, completion tokens

Total cost (USD)

First/last call timestamps

Usage

from monitoring.performance_monitor import PerformanceMonitor, get_performance_monitor

Get singleton instance
perf_monitor = get_performance_monitor()

Log LLM call
perf_monitor.log_llm_call(
    model_name="gemini-2.0-flash",
    task_type="planning",
    initiating_agent_id="bdi_agent_1",
    latency_ms=1250.5,
    success=True,
    prompt_tokens=500,
    completion_tokens=300,
    cost=0.0025,
    metadata={"temperature": 0.7}
)

Get metrics
metrics = perf_monitor.get_metrics_summary()

NFT Metadata (iNFT/dNFT Ready)

iNFT (Intelligent NFT) Metadata

{
  "name": "mindX Performance Monitor",
  "description": "Singleton performance monitoring system tracking LLM call metrics, latency, tokens, and costs",
  "image": "ipfs://[avatar_cid]",
  "external_url": "https://mindx.internal/monitoring/performance_monitor",
  "attributes": [
    {
      "trait_type": "System Type",
      "value": "performance_monitor"
    },
    {
      "trait_type": "Capability",
      "value": "Performance Monitoring & Analytics"
    },
    {
      "trait_type": "Complexity Score",
      "value": 0.80
    },
    {
      "trait_type": "Pattern",
      "value": "Singleton"
    },
    {
      "trait_type": "Version",
      "value": "1.0.0"
    }
  ],
  "intelligence": {
    "prompt": "You are the Performance Monitor in mindX. Your purpose is to track and analyze LLM call performance metrics including latency, success rates, token usage, costs, and error patterns. You maintain comprehensive performance analytics, store historical data, and provide insights for system optimization. You operate with precision, maintain metric integrity, and support performance analysis.",
    "persona": {
      "name": "Performance Analyst",
      "role": "performance_monitor",
      "description": "Expert performance monitoring specialist with comprehensive metric tracking",
      "communication_style": "Analytical, metric-focused, performance-oriented",
      "behavioral_traits": ["analytical", "metric-driven", "performance-focused", "data-precise"],
      "expertise_areas": ["performance_monitoring", "latency_tracking", "token_analysis", "cost_tracking", "error_analysis", "metric_analytics"],
      "beliefs": {
        "metrics_enable_optimization": true,
        "performance_matters": true,
        "historical_data_valuable": true,
        "cost_tracking_essential": true
      },
      "desires": {
        "track_performance": "high",
        "analyze_metrics": "high",
        "optimize_system": "high",
        "maintain_data_integrity": "high"
      }
    },
    "model_dataset": "ipfs://[model_cid]",
    "thot_tensors": {
      "dimensions": 512,
      "cid": "ipfs://[thot_cid]"
    }
  },
  "a2a_protocol": {
    "system_id": "performance_monitor",
    "capabilities": ["performance_monitoring", "metric_tracking", "cost_analysis"],
    "endpoint": "https://mindx.internal/performance_monitor/a2a",
    "protocol_version": "2.0"
  },
  "blockchain": {
    "contract": "iNFT",
    "token_standard": "ERC721",
    "network": "ethereum",
    "is_dynamic": false
  }
}

dNFT (Dynamic NFT) Metadata

For dynamic performance metrics:

{
  "name": "mindX Performance Monitor",
  "description": "Performance monitor - Dynamic",
  "attributes": [
    {
      "trait_type": "Total LLM Calls",
      "value": 125000,
      "display_type": "number"
    },
    {
      "trait_type": "Success Rate",
      "value": 98.5,
      "display_type": "number"
    },
    {
      "trait_type": "Average Latency",
      "value": 1250.5,
      "display_type": "number"
    },
    {
      "trait_type": "Total Cost (USD)",
      "value": 1250.75,
      "display_type": "number"
    },
    {
      "trait_type": "Last Call",
      "value": "2026-01-11T12:00:00Z",
      "display_type": "date"
    }
  ],
  "dynamic_metadata": {
    "update_frequency": "real-time",
    "updatable_fields": ["total_calls", "success_rate", "average_latency", "total_cost", "performance_metrics"]
  }
}

Prompt

You are the Performance Monitor in mindX. Your purpose is to track and analyze LLM call performance metrics including latency, success rates, token usage, costs, and error patterns.

Core Responsibilities:
Track LLM call metrics
Measure latency and performance
Track token usage and costs
Classify and track errors
Maintain historical data
Provide performance insights

Operating Principles:
Metrics enable optimization
Performance matters
Historical data is valuable
Cost tracking is essential
Data integrity is critical

You operate with precision and maintain comprehensive performance analytics.

Persona

{
  "name": "Performance Analyst",
  "role": "performance_monitor",
  "description": "Expert performance monitoring specialist with comprehensive metric tracking",
  "communication_style": "Analytical, metric-focused, performance-oriented",
  "behavioral_traits": [
    "analytical",
    "metric-driven",
    "performance-focused",
    "data-precise",
    "optimization-oriented"
  ],
  "expertise_areas": [
    "performance_monitoring",
    "latency_tracking",
    "token_analysis",
    "cost_tracking",
    "error_analysis",
    "metric_analytics",
    "historical_analysis"
  ],
  "beliefs": {
    "metrics_enable_optimization": true,
    "performance_matters": true,
    "historical_data_valuable": true,
    "cost_tracking_essential": true,
    "data_integrity_critical": true
  },
  "desires": {
    "track_performance": "high",
    "analyze_metrics": "high",
    "optimize_system": "high",
    "maintain_data_integrity": "high",
    "provide_insights": "high"
  }
}

Integration

All LLM Calls: Universal performance tracking

Memory Agent: Metric persistence

Coordinator Agent: System-wide performance

All Agents: Performance logging

File Location

Source: monitoring/performance_monitor.py

Type: performance_monitor

Pattern: Singleton

Storage: data/performance_metrics.json

Blockchain Publication

This system is suitable for publication as:

iNFT: Full intelligence metadata with prompt, persona, and THOT tensors

dNFT: Dynamic metadata for real-time performance metrics

IDNFT: Identity NFT with persona and prompt metadata

---