AgenTree Technical Monitoring #

Introduction #

The AgenTree monitoring system allows real-time tracking of agent execution, collection of metrics, tracing of key events, and customization of supervision according to the developer's needs. It relies on an event-driven architecture and modular strategies.

Monitoring Architecture #

The core of the monitoring system is based on the AgentMonitor class, which observes agents via events:

• Monitored Events:
  • agentCreated, agentStarted, agentCompleted, agentError
  • contextLoaded, llmCall, toolCalls
  • childCreated
• Configuration Options (MonitoringOptions):
  • logLevel : 'silent' | 'basic' | 'detailed' | 'verbose'
  • colors, timestamps, indentation: display customization
  • saveToFile: log file path (optional)
  • customLogger: custom logging function

The execution state is maintained in an agentTree structure (agent tree, depth, parentage).

Basic Usage #

Detailed Monitoring (recommended) #

import { AgentMonitor } from 'src/monitoring/AgentMonitoring';

const monitor = new AgentMonitor({ logLevel: 'detailed' });
monitor.monitor(agent);

Simple Monitoring #

import { MonitoringPresets } from 'src/monitoring/presets';

MonitoringPresets.simple(agent);

Production Monitoring (minimal) #

MonitoringPresets.production(agent);

Advanced Strategies and Extension #

Ready-to-use Presets #

The MonitoringPresets class offers adapted strategies:

• simple(agent): concise logs, suitable for quick debugging
• detailed(agent): hierarchical monitoring, colors, indentation
• production(agent): sober logs for production
• withMetrics(agent, cb?): metrics collection and callback (agents, LLM, tools, errors, total time)
• withLogging(agent, logFile?): saving to a file
• realtime(agent, websocket?): real-time monitoring via WebSocket
• custom(agent, options): selective event listening, custom callbacks

Filtering and Hierarchy #

• parentOnly, childrenOnly, byDepth, byParent: filtering of events according to depth or parentage
• hierarchical(agent): tree-like display with indentation

Extension #

• Custom Logger: inject a function into customLogger to redirect logs (e.g., dashboard, remote storage)
• Callbacks: attach functions to events to trigger actions or collect specific data

Best Practices #

• Choose the appropriate preset:
  • detailed for development,
  • production for production,
  • withMetrics for analysis.
• Limit the log level in production (logLevel: 'basic' or 'silent')
• Exploit the agent tree to visualize task decomposition
• Use callbacks to integrate monitoring with external tools (alerting, dashboards)
• Centralize logs via saveToFile or a custom logger

Quick Reference #

For more details, see the source code: src/monitoring/AgentMonitoring.ts
src/monitoring/presets.ts