High-precision telemetry and agent health metrics
Monitor agent fleet health with real-time aggregated metrics: token generation rates, model latency percentiles, tool execution durations, failure rates, and cache hit ratios on customizable dashboards.
// Real-time aggregate metrics exported via OpenTelemetry
// Ingested & aggregated every 5 seconds
const agentMetrics = {
p50_latency_ms: 640,
p95_latency_ms: 1890,
p99_latency_ms: 3420,
tokens_per_sec: 142.8,
error_rate_pct: 0.12,
cache_hit_rate: 0.84, // Anthropic / OpenAI prompt cache
};Engineered for high-throughput autonomous agents
Every capability is built into the OpenTelemetry streaming pipeline with sub-millisecond ingestion overhead.
Model & Agent Latency Percentiles
Analyze how LLM inference times fluctuate across providers (OpenAI, Anthropic, Gemini, Groq) and identify slow tool bottlenecks.
- ✓P50, P90, P95, and P99 latency percentiles over time
- ✓Time-to-first-token (TTFT) vs total generation time
- ✓Tool API latency vs LLM reasoning latency breakdown
Token Throughput & Capacity Monitoring
Track prompt tokens, output tokens, reasoning tokens, and caching efficiency across projects and agent teams.
- ✓Tokens per second (TPS) generation velocity
- ✓Anthropic prompt cache & OpenAI prefix cache hit rates
- ✓Provider rate limit threshold alerting (TPM / RPM)
Automated Error & Anomaly Alerting
Detect unexpected error spikes, infinite tool call loops, model hallucination surges, or network timeouts immediately.
- ✓Configurable thresholds for error rates and latency degradation
- ✓Webhook, Slack, Discord, and PagerDuty alert integrations
- ✓Root-cause trace link included with every triggered alert
Frequently Asked Questions
Can I create custom metrics for my business logic?▼
Can metrics be integrated with Grafana or Datadog?▼
Related Platform Features & Guides
Ready to monitor and secure your AI agents?
Get started in under 3 minutes with zero credit card required. Free tier includes up to 5 projects and community telemetry.