NVIDIA Fleet Intelligence Host Agent

NVIDIA Fleet Intelligence Host Agent collects GPU health, system telemetry, machine information, and attestation data from Linux hosts. Infrastructure operators can use it to monitor GPU nodes and investigate hardware and system problems.

Features and use cases

• Collect GPU metrics, including power, temperature, utilization, and memory usage.
• Collect CPU, memory, disk, and network telemetry alongside GPU health and events.
• Expose data through HTTP and Prometheus endpoints, write local files, or export through OTLP.
• Deploy on bare-metal hosts or Kubernetes GPU nodes.

Requirements

Use a supported Linux host and the appropriate image architecture (AMD64 or ARM64). GPU collection requires supported NVIDIA GPUs, NVIDIA drivers, access to host GPU resources, and a reachable DCGM HostEngine. Persist /var/lib/fleetint to retain node identity and enrollment state. Remote export requires network access and any credentials required by the configured backend.

Documentation

See the project README, configuration guide, and Kubernetes installation guide for supported platforms and deployment instructions.

Publisher and source

Published by NVIDIA. Source code and contribution history are available in the Fleet Intelligence Agent repository.

License

The Fleet Intelligence Agent project source is licensed under the Apache License 2.0. Container dependencies retain their respective licenses.