Performance Metrics

Metrics available in LeafMesh for monitoring system health and performance.

Built-in Metrics

Agent Health Metrics

LeafMesh tracks per-agent metrics automatically and surfaces them in the dashboard:

Metric	Description
`response_time_ms`	Average response time
`error_rate`	Error ratio
`success_rate`	Success ratio
`active_sessions`	Current sessions
`consecutive_failures`	Failures without success
`uptime_seconds`	Time since last start
`quality_score`	Output quality

LLM Cost Metrics

Every LLM call contributes cost data — model, prompt/completion tokens, estimated cost, response time, cache hit/miss — which is aggregated in the dashboard and accessible via:

Aggregated analytics, cache metrics, and per-session event streams are available in Studio's Dashboard and session inspector — operator UI, not part of the customer Python or HTTP surface.

Event Processing Metrics

Aggregated analytics, cache metrics, and per-session event streams are available in Studio's Dashboard and session inspector — operator UI, not part of the customer Python or HTTP surface.

Metric	Description
`total_events`	Total events processed
`events_by_type`	Breakdown by event type
`processing_rates`	Events/second

Performance Baselines

Typical performance characteristics:

Operation	Expected Latency
Request routing	10-50ms
Condition evaluation	~1ms
LLM call (fast model)	300-1000ms
LLM call (capable model)	500-5000ms
Tool execution	Varies
Mesh call overhead	5-20ms

The LLM call dominates request latency; framework overhead is small.

Custom Metrics

Build custom metrics using the public APIs:

async def collect_custom_metrics():
    analytics = Studio's Dashboard (HTTP)
    agent_stats = leafmesh.get_agent_stats()
    cache_stats = Studio's Dashboard (HTTP)

    return {
        "total_events": analytics.get("total_events", 0),
        "error_rate": analytics.get("error_rate", 0),
        "agents": agent_stats,
        "cache_hit_rate": cache_stats.get("hit_rate", 0),
        "estimated_savings": cache_stats.get("estimated_savings", 0)
    }

Adaptive Model Selection Metrics

When adaptive model selection is enabled, LeafMesh tracks per-model history (response time, output quality, cost per request, success rate, throughput) and uses it to inform future routing decisions.

Next Steps

Dashboard — Building monitoring dashboards
Alerting — Alert configuration
Monitoring — System monitoring overview

LeafMesh — Performance metrics reference