---
title: "GenAI Agent Observability | Grafana Cloud documentation"
description: "Monitor LLM performance, token usage, costs, and user interactions"
---

> For a curated documentation index, see [llms.txt](/llms.txt). For the complete documentation index, see [llms-full.txt](/llms-full.txt).

# GenAI Agent Observability

GenAI Agent Observability provides comprehensive monitoring for AI agent systems including invocation tracking, cost analysis, performance metrics, and operational insights across your agentic AI applications.

## Overview

The Agent Observability dashboard monitors AI agent applications, offering insights into:

- **Invocation monitoring** - Total invocations, distribution by source, and usage patterns
- **Cost analysis** - Real-time spend tracking and per-agent cost breakdown
- **Performance analytics** - Operation duration, latency percentiles, and throughput rates
- **Provider insights** - Performance comparison across LLM providers
- **Operational logs** - Agent interaction logs with distributed tracing correlation

## Key features

This dashboard provides panels for invocation traction, cost management, performance monitoring, and logs and debugging.

### Invocation tracking

- Total agent invocation volume and frequency tracking
- Invocation distribution by agent source
- Percentage breakdown across agent types
- Usage pattern identification and trend analysis

### Cost management

- Real-time total agent cost tracking in USD
- Per-agent cost breakdown and attribution
- Cost comparison across different agents
- Spend visibility across time ranges and environments

### Performance monitoring

- 95th percentile (p95) operation duration by agent
- Heatmap visualization of latency distribution over time
- Average operation duration by agent and LLM provider
- Operation throughput rate (requests per second)

### Logs and debugging

- Integrated agent interaction logs
- Agent name filtering for targeted debugging
- Contextual log output with agent name and message details
- Filter capabilities for targeted debugging and root cause analysis