Metrics & Analytics

Real-time latency, tokens, cost, and success rate monitoring.

Avg Latency

1434ms

13:44:4013:46:15

Tokens / Request

1,838

13:44:4013:46:15

Success Rate

95.5%

13:44:4013:46:15

Cost / Request

$0.0100

13:44:4013:46:15

Model Breakdown

Model	Calls	Avg Latency	Tokens	Cost	Success
gpt-4o	1,247	1420ms	2.5M	$14.70	98.2%
gpt-4o-mini	3,891	380ms	5.1M	$3.07	99.1%
claude-3.5-sonnet	892	1180ms	1.8M	$8.01	97.8%

Tool Performance

Tool	Calls	Avg Latency	Success	Errors
web-search	2,340	1850ms	94.2%	136
calculator	1,580	12ms	99.9%	2
code-exec	890	450ms	96.5%	31
db-query	670	85ms	98.1%	13

Integration Code

import { createMetrics, PrometheusExporter, DatadogExporter } from 'agent-tools-kit/observability'

const metrics = createMetrics({
  exporters: [
    new PrometheusExporter({ port: 9090 }),
    new DatadogExporter({ apiKey: process.env.DD_API_KEY }),
  ],
  // Built-in metric collectors
  collect: ['latency', 'tokens', 'cost', 'success_rate', 'tool_usage'],
  // Custom dimensions
  dimensions: ['model', 'tool', 'agent_id', 'user_tier'],
  // Alerting rules
  alerts: [
    { metric: 'latency_p99', threshold: 5000, action: 'pagerduty' },
    { metric: 'success_rate', below: 95, action: 'slack', channel: '#agent-alerts' },
    { metric: 'cost_per_hour', threshold: 50, action: 'email' },
  ]
})

agent.use(metrics.middleware())

// Custom metric tracking
metrics.record('custom_score', 0.95, { model: 'gpt-4o', task: 'summarization' })

// Query metrics programmatically
const p99 = await metrics.query('latency', { percentile: 99, window: '1h' })