Skip to main content

Blog post · November 19, 2026 · aviznetworks.com

Aviz Networks details Network Copilot's Prometheus agent

Summary

Aviz Networks describes a Network Copilot Prometheus agent that reads the metrics a deployment holds and answers questions about them without predefined setup. The approach lets enterprise IT teams query new metric sources without configuration changes.

Supported metric sources

Network Copilot reads whatever Prometheus holds, regardless of how a metric arrived. The stated sources are standard exporters for nodes, GPUs, inference servers, Kubernetes and networks, custom exporters from internal services and tools, and the Pushgateway for batch jobs and short-lived tasks that cannot be scraped. According to the post, no registration step is needed in Copilot for any of these sources.

Metric interpretation

The agent looks up the metrics and labels present at query time rather than working from a fixed list. It draws meaning from each metric's description, type, units and labels, converting counters to rates and respecting units. Metrics that share node, GPU, pod, interface or endpoint labels can be joined and compared on the same timeline.

Dashboards and example queries

Users can request a view, and Copilot assembles charts, tables and KPI summaries in the conversation, with changes such as splitting by node applied on request. The post cites examples including a cache hit ratio query on a newly added exporter, a nightly InfiniBand and NVLink check pushed to the Pushgateway, and a single-node health view covering GPU, host and interface signals. In the inference example, the agent reports a sequence running from GPU memory pressure through cache saturation and queuing to latency, along with when it started.

Source attribution and access

According to the post, every number comes from a query run in the conversation, and its source is displayed. The agent checks Prometheus before reporting that a metric is not collected, and states which interpretation it applied when a metric can be read more than one way. Each project queries only its own Prometheus instance under its users' permissions.

Network Copilot's Prometheus agent, as described by Aviz Networks, is designed to work across scraped, custom and pushed metrics without per-metric configuration. The post presents this as removing the need to wait for dashboards or rules when new metrics are added, which is relevant to enterprise teams managing varied telemetry sources. This Blog Summary is a fact-based summary of the vendor blog.

Our summary. The original keeps its own byline.

Read the original on aviznetworks.com

Companies in this article

Topics