# Agents Monitoring

> Monitor LLM and agent usage across your applications — conversations, token consumption, latency, cost, success rates, and active models — built on OpenTelemetry GenAI semantic conventions

Source: https://last9.io/docs/agents-monitoring/

Agents Monitoring gives you a single view of how your applications use large language models and AI agents: how many conversations they run, what they cost, how long calls take, how often they fail, and which models are actually in use.

![Agents Monitoring overview](../../../../assets/content/docs/discover/llm/agents-monitoring-overview.png)

The view is built on the [OpenTelemetry GenAI semantic conventions](https://opentelemetry.io/docs/specs/semconv/gen-ai/). Any instrumentation that emits GenAI-convention telemetry feeds it — no Last9-specific SDK required.

## Overview

Open **Agents Monitoring** under the AI section of the sidebar. The Overview tab reports, for the selected environment and time range:

| Card                  | What it shows                                                      |
| --------------------- | ------------------------------------------------------------------ |
| **Conversations**     | Total conversations and average messages per conversation          |
| **Conversation Cost** | Average cost per conversation                                      |
| **Latency**           | P50, P90, P95, and P99 call duration                               |
| **Total LLM cost**    | Spend across all calls in the range                                |
| **Spans**             | Total LLM spans recorded                                           |
| **Token count**       | Total tokens consumed                                              |
| **Success rate**      | Success and error rate across all requests                         |
| **Top Models**        | Models serving traffic, with request counts and share              |
| **Model Insights**    | Overall performance, cost efficiency, token usage, model diversity |

The **Conversations** tab lists individual conversations, and the **Traces** tab shows the underlying LLM spans — a slow or failing call links to the full request trace: the endpoint that triggered it, retries, and downstream work.

## Metrics

The overview is backed by GenAI semantic-convention metrics in your Last9 workspace — `gen_ai_client_token_usage_total` for tokens and the `gen_ai_client_operation_duration_seconds` histogram for request counts and latency, with `gen_ai_request_model` and `error.type` labels. The same series are queryable with PromQL in dashboards, alerts, and ad-hoc queries.

## Getting Data In

Instrument your application with any library that emits GenAI semantic-convention telemetry. See the [AI integrations](/docs/integrations/) for supported options, including the [Python GenAI SDK](/docs/integrations/python-genai-sdk/) for LLM observability in Python applications.

:::note
Monitoring for coding agents (Claude Code, Cursor, Codex, and others) is a separate **Coding Agents** view fed by the agents' own OpenTelemetry output. See the [coding agent integrations](/docs/integrations/) to connect them.
:::

---

## Troubleshooting

- **The overview is empty**: no GenAI-convention telemetry has arrived yet. The empty state links to the integrations to set one up. Verify your instrumentation exports `gen_ai_*` metrics by checking the metrics explorer for `gen_ai_client_token_usage_total`.

Please get in touch with us on [Discord](https://discord.com/invite/Q3p2EEucx9) or [Email](mailto:support@last9.io) if you have any questions.