NVIDIA Config Manager
docs.nvidia.com
- Free plan
- Linux
- Web
- API
- Self-hosted
What NVIDIA Config Manager is
NVIDIA Config Manager automates the setup and day-to-day operation of NVIDIA AI-factory network infrastructure. It is deployed with Helm on Kubernetes, using an installer that supports an interactive TUI or a headless CLI. Documented workflows cover configuration backups and deployment, cable and hardware validation, password rotation, reprovisioning, and operating system or firmware upgrades. The platform combines workflow orchestration and template rendering with PostgreSQL-backed configuration storage, zero-touch provisioning, and Kea DHCP configuration generation. Nautobot is the default DCIM provider; other providers can connect through the provider-neutral SDK. Browser authentication uses an OIDC proxy, while API clients use bearer tokens. Approval workflows and an audit trail support day-two changes. Cloud-hosted and on-premises Kubernetes deployments are documented, including an airgapped option. Standard deployment requires Kubernetes access, kubectl, Helm 3.x, Python 3.13 or later, and uv. The repository describes the software as in Developer Preview and not recommended for production use. The listed Apache 2.0 open-source plan is free.
Who it suits
It suits network infrastructure administrators deploying and maintaining the platform, and operators running network workflows. Its Developer Preview status makes it a poor fit for production use, according to the repository.
Good to know, and what to watch
- Automates documented network provisioning and maintenance workflows.
- Supports cloud-hosted and on-premises Kubernetes deployments.
- Offers an airgapped deployment bundle.
- Approval workflows and audit trail support changes.
- Includes zero-touch provisioning and API access.
- Developer Preview; not recommended for production use.
- Requires Kubernetes, kubectl, Helm 3.x, Python 3.13+, and uv.
- Large production or high-availability profile specifies 96 vCPUs and 256 GB RAM.
Plans
Every plan and priceApache 2.0 open-source license
Open source
The verdict
Config Manager brings provisioning, configuration workflows, and change controls together for NVIDIA AI-factory networks. Its preview status and Kubernetes requirements are central considerations before deployment.
What the maker publishes
- Authentication
- Browser authentication uses an OIDC proxy, while API clients can authenticate with bearer tokens. source
- Change controls
- The platform supports approval workflows for day-two changes and documents an audit trail for configuration changes. source
- Core services
- The platform includes workflow orchestration, template rendering, PostgreSQL-backed configuration storage, zero-touch provisioning, and Kea DHCP configuration generation. source
- DCIM integration
- Nautobot is the bundled default DCIM provider, and other providers can integrate through the provider-neutral DCIM SDK. source
- Deployment
- The platform is deployed with Helm on a Kubernetes cluster, with an installer offering interactive TUI and headless CLI workflows. source
- Hosting
- Documented deployment models include cloud-hosted and on-premises Kubernetes deployments. source
- Intended users
- The documented primary personas are network infrastructure administrators who deploy and maintain the platform and operators who run workflows and manage network operations. source
- Network operations
- Documented workflows include configuration backup and deployment, cable and hardware validation, password rotation, reprovisioning, and operating system or firmware upgrades. source
- Offline deployment
- An airgapped deployment bundle can include the Helm chart, container images, dependency charts and manifests, and an offline installer package. source
- Preview status
- The repository describes Config Manager as in Developer Preview and not recommended for production use. source
- Purpose
- NVIDIA Config Manager automates bootstrapping and operation of NVIDIA AI-factory network infrastructure. source
- Requirements
- A standard deployment requires Kubernetes access, kubectl, Helm 3.x, Python 3.13 or later, and uv; some deployment workflows also use Docker and Kind. source
- Scale
- The documented large sizing profile for production or high-availability validation specifies 96 total vCPUs, 256 GB total RAM, and at least three nodes. source
- Support and security reporting
- The repository links to a support policy and instructs users to report security vulnerabilities through its private disclosure process rather than public GitHub issues. source
Checked against the maker's pages on 4 Oct 2026.