See it. Plan it. Boot it.
A bare-metal operations console for NVIDIA Infra Controller.
English · 简体中文
Features · Getting started · User guides · Architecture · Documentation
PXEye brings device discovery, inventory, address planning and operating-system installation into one interface. Built on NVIDIA Infra Controller (NICo), it helps infrastructure teams move from newly arrived hardware to verified OS delivery, with explicit execution steps and persistent installation records.
PXEye is an independent community project. NICo provides the infrastructure control plane; PXEye provides the operator interface, planning resources and workflow orchestration.
An illustrative management view inspired by GB200 NVL72, showing a subset of compute, NVLink and power equipment alongside management servers. Identifiers, addresses and task states are authored examples. The interface currently uses Simplified Chinese.
- Understand the hardware. Identify devices by serial number, inspect management endpoints and organize equipment by rack.
- Plan before executing. Save addresses, target systems and disk policies independently of hardware changes. Apply them through explicit operations.
- Follow installation to completion. Keep the effective configuration, events, console logs and acceptance results with each task. Retain failure evidence for recovery.
| Area | What you can do |
|---|---|
| Rack topology | View servers, DPU endpoints, NVLink switches and power shelves; persist manual device ordering. |
| Discovery and planning | Import arrival CSV plans separately; discover DHCP endpoints, verify identity and enroll devices, then match address configuration by unique SN. |
| Address management | Maintain planned BMC and host management addresses; update supported controllers and refresh observed details in batches. |
| Networks | Inspect NICo network segments, create installation subnets and review reservations and DHCP prerequisites. |
| Operating systems | Upload or import Ubuntu ISOs from a service-accessible directory, extract boot artifacts, and register NICo OS definitions. |
| Installation profiles | Reuse disk-selection, GPT/UEFI partitioning, ext4/XFS and installation defaults; freeze resolved device settings when creating tasks. |
| Installation tasks | Prepare hardware when needed, allocate or reuse Instances, launch unattended installation, and inspect logs and OS acceptance results. |
| NICo cluster | Deploy online Charts or uploaded offline material, inspect service status and connect site settings. |
| Instances and site | Inspect and release Instances; view Core connectivity, client identity, storage and supporting services. |
See Features and support boundaries for native resource mappings and hardware requirements.
PXEye can start independently and use the NICo cluster page to deploy the control plane in its current Kubernetes cluster, or connect to existing NICo services. Prepare the control plane and installation networks, verify one server, then expand the batch.
- An existing Kubernetes cluster, compatible client tools and persistent PXEye storage are available; choose an online Chart source or upload offline material and provide its values; use Core mTLS credentials for an existing site.
- Target server serial numbers, CPU architectures and permanent BMC/PXE MAC addresses are verified.
- The actual control-plane path can verify BMC Redfish identity and permissions, with the credentials required by the selected ingestion workflow.
- BMC and host boot networks have agreed subnets, gateways, VLANs, relays and conflict-free address allocations.
- The actual installation path supports DHCP requests and replies plus the required HTTP, Core, DNS and NTP services.
- Matching EFI, Scout, ISO, kernel/initrd, unattended-install settings and software sources are prepared for the target architecture.
- RAID arrangements, stable system-disk selection rules, backups and permitted cleanup scope are confirmed.
- Artifact mounts and publication URLs are configured alongside OS login, installer callbacks and acceptance access.
Use the full pre-installation checklist for the discovery, OS-writing and managed-DPU stages, with verification references and handoff fields.
For a new control plane, follow Start PXEye and deploy NICo. For an existing NICo site, follow the deployment guide with the site configuration and single-replica manifest. Configure reachable installation network entry points in either case.
Compatible nico-admin-cli, grpcurl and Helm 3 binaries are build inputs; they are not bundled in the source repository. Build commands and mount locations are maintained in the deployment guide.
Follow Install your first server on a dedicated device: prepare its OS, identify and enroll it, save its plan, then install and verify system-disk boot, SSH and a normal reboot. Retain the task evidence before expanding the batch.
- Prepare the site: verify NICo connectivity, DHCP relay, network segments and artifact delivery.
- Prepare an OS: import an ISO, prepare boot artifacts and register its NICo definition.
- Identify and enroll: import arrival plans, discover DHCP endpoints, retrieve hardware identity, enroll devices and match plans by SN.
- Save the plan: choose addresses, the target OS, an installation profile and device overrides.
- Install and verify: create a task, review its target and disk scope, then follow execution through SSH and normal-reboot verification.
Discovery, planning and installation have separate responsibilities. Non-server equipment uses the inventory and address workflow; PXEye does not install its OS.
These images share the same illustrative dataset. The reference scenario explains its scope and how to reproduce the captures.
PXEye uses a Python HTTP backend, SQLite/WAL persistence and native JavaScript/CSS. The service serves both the UI and API; no frontend build step is required. NICo Core owns infrastructure lifecycle state. PXEye stores device plans, profiles, task snapshots, logs and display layout.
The current technical preview targets a single site, one service replica and a single operator, with configurable installation concurrency within a batch (default: 5). The principal physical validation path is Ubuntu Server 24.04 on x86_64 UEFI hardware. ARM64, GB200/GB300 and other OEM workflows require separate validation. Address operations depend on each device's Redfish or NVOS capabilities.
Kubernetes provisioning, workload scheduling, firmware distribution, fabric configuration and multi-tenant RBAC are outside the current product scope. See Architecture and Security.
All public project documentation is available in English and Simplified Chinese. Start with the documentation index for operational guides, design references and contribution information. The changelog records capability milestones.
Browse by menu for purpose, field values and actions. See the user guide index for the complete list.
| Page | Guide |
|---|---|
| Rack topology | Rack groups, device lookup and ordering |
| Tasks | Select devices, install and inspect results |
| Devices | Plan import, enrollment, addresses and system configuration |
| Instances | Inspect instances and reclaim devices |
| Operating systems | Import ISOs, prepare and enable OS versions |
| Networks | Subnet creation, reservations and DHCP prerequisites |
| Installation profiles | OS, disk, partition and login defaults |
| Site settings | Connections, storage and installation concurrency |
| NICo cluster | Deploy services and connect a site |
See CONTRIBUTING.md for development principles and change-review expectations. Include the target NICo version, device capabilities and validation scope when proposing a change. Documentation improvements and hardware compatibility reports are welcome.
PXEye is a technical preview licensed under Apache-2.0. Dependencies and third-party assets retain their own licenses. See Security for the current reporting process.


