GB10 Manager v1.0.0

GB10 Manager is a web application for system and application management designed specifically for the NVIDIA GB10 Grace Blackwell. It provides a modern and intuitive interface to monitor, configure and deploy services on the ARM64 platform.

Main interface with System tab active

Features

Architecture

GB10 Manager is a Go application with an embedded web frontend, deployed as a Docker container on the GB10.

Architecture Diagram

Technical Stack

Prerequisites

• NVIDIA GB10 Grace Blackwell with Ubuntu 24.04 LTS
• Docker Engine installed and running
• NVIDIA Container Toolkit for GPU access (--gpus all)
• Ollama installed (native or Docker container) on port 11434
• Caddy (optional) for HTTPS reverse proxy
• SSH access to the GB10 server

Graphical Installer

The gb10-installer.exe installer is a standalone Windows binary that deploys GB10 Manager automatically to a remote GB10 server via SSH.

Usage

1. Copy the releases/ folder to your Windows machine
2. Launch gb10-installer.exe - a web interface opens in the browser
3. Step 1 - Connection: Enter the IP address, SSH port (22), username and password
4. Step 2 - Configuration: Choose the destination directory (/opt/gb10-manager by default)
5. Step 3 - Installation: Follow progress in real-time (SSE streaming)
6. Step 4 - Models: Select and download desired Ollama models

Releases directory contents

releases/
  gb10-installer.exe     # Windows installer (UPX compressed)
  payload/
    gb10-manager         # ARM64 Linux binary (UPX compressed)
    Dockerfile           # Dockerfile for the gb10-manager container
    docs.tar.gz          # Web documentation (gb10-docs container)

Manual Installation

1. Copy files to the server

scp gb10-manager-arm64 eforgues@192.168.1.89:/opt/gb10-manager/gb10-manager
scp Dockerfile eforgues@192.168.1.89:/opt/gb10-manager/Dockerfile

2. Build Docker image

cd /opt/gb10-manager
docker build --no-cache -t gb10-manager:latest -f Dockerfile .

3. Start the container

docker run -d \
  --name gb10-manager \
  --network host \
  --gpus all \
  --restart unless-stopped \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -v /etc:/host-etc:ro \
  gb10-manager:latest

Docker flags explained

Header and Navigation

The header displays the server name (GB10 hostname), Ollama status in real-time and access to settings. Main navigation uses four tabs.

Header with server name, Ollama status, tabs and quick links

Header Components

• Server name: Displays the GB10 hostname (Orbitron font, blue gradient). Dynamically fetched via /api/system/info.
• Ollama indicator: Green (online) or red (offline) dot with status text. Periodic check via /api/ollama/diagnostic.
• Settings button: Gear icon opening the settings modal.

Quick Links Bar

Automatically displays links to detected Docker services. Each link points to https://hostname:port and opens in a new tab. Recognized services include: Open WebUI, Whisper ASR, DocMind AI, Stirling PDF, Obsidian and any application configured with an exposed port.

System Tab

The main tab displays GB10 system information, real-time statistics and the complete Ollama diagnostic.

System Information Cards

Cards display the server hardware and software information:

GB10 system information cards

Real-time Statistics

Statistics are automatically refreshed every 3 seconds via the endpoint GET /api/system/stats. Each metric is shown with a progress bar and a numeric value.

Real-time metrics: CPU, GPU, RAM, temperatures, power and frequencies

Dummy / Expert Mode

The Dummy mode (by default) blocks execution of potentially dangerous commands. The Expert mode unlocks all commands and enables the SSH Terminal button which opens an interactive WebSSH terminal to the GB10.

Ollama Diagnostic

The Ollama Diagnostic section provides a complete view of the Ollama service status, consumed resources, loaded models in memory and installed models. It also allows downloading or deleting models.

Complete Ollama diagnostic: container, system, models and download

Diagnostic Features

• Container stats: CPU, RAM, memory limits, network and disk I/O, PID count
• System stats: Total/used/available RAM, swap, GPU (name, utilization, temperature, VRAM)
• Loaded models: List of models currently in GPU memory with expiration time
• Installed models: All downloaded models with size and delete button
• Download a model: Input field with real-time SSE progress (bar + percentage)

Ollama APIs Used

Docker Containers

The Containers tab displays all Docker containers on the system with their status, resource usage and quick actions (start, stop, restart).

Docker containers list with status and statistics

Displayed Information

• Name: Docker container name
• Uptime: Duration since startup or last exit
• Resources: CPU and RAM usage (auto-refresh)
• Status: Colored badge (running = green, stopped = red)
• Actions: Start, stop or restart the container (Expert mode)

Applications

The Applications tab offers an advanced management interface for deployed services. It includes a sidebar with the list of detected applications and a detail panel with information, logs and available actions.

Application management interface with service details

Supported Applications (36 services)

GB10 Manager automatically recognizes 36 pre-configured Docker services, including:

... and 26 more services (PostgreSQL, Redis, MinIO, Milvus, Home Assistant, etc.)

Installing a New Application

The + Install button opens a modal to specify the application name, installation method (Docker, APT, Snap) and version. The Ask AI button uses Ollama to generate optimal installation commands.

AI Chat (Ollama)

The chat panel at the bottom of the page allows interacting with Ollama for help with installation, configuration and troubleshooting. Commands suggested by the AI can be executed directly after validation.

AI Chat with Ollama - interactive assistance

Chat Features

• SSE Streaming: Responses arrive word by word in real-time
• History: Conversation history is preserved during the session
• Command execution: Commands in AI suggestions can be executed with one click (with validation)
• Configurable model: The Ollama model used is configurable in settings
• Collapsible panel: The chat can be shown/hidden via the toggle in options

Settings - General

The settings modal is accessible via the gear icon in the header. The General tab contains language, execution mode, persistent services and system information.

General settings tab

Settings - Options

The Options tab controls interface behavior: AI chat, default Ollama model and compliment popup on startup.

Chat options, AI model and popup settings

Settings - Themes

GB10 Manager supports 5 visual themes based on the Catppuccin. Each theme changes the background, text, accent, success, warning and error colors.

Catppuccin theme selection

Available Themes

Settings - System

The System tab allows modifying GB10 system settings: hostname, timezone, DNS, locale. Each change is analyzed by AI before application to ensure system stability.

Modifiable Settings

Settings - HTTPS Certificate

The Certificate tab manages SSL certificates generated by Caddy for HTTPS.

Features

• Caddy Status: Shows if Caddy is running
• Certificate Info: Active certificate details (issuer, validity, subject)
• CA Download: Download the CA certificate to install in your browser or operating system
• Renewal: Deletes and regenerates Caddy's internal PKI. All clients must reinstall the new CA certificate after renewal

API Reference

GB10 Manager exposes a REST API on port 8100. All endpoints return JSON unless otherwise specified.

System

Ollama

Containers

Applications

Chat

Services

Whisper

Certificates

Healthcheck

Example de requete

Download a model Ollama (SSE)
curl -N -X POST http://localhost:8100/api/ollama/models/pull \
  -H "Content-Type: application/json" \
  -d '{"name":"llama3.1:8b"}'

# Reponse SSE (flux continu) :
data: {"status":"pulling manifest"}
data: {"status":"downloading","completed":1234567,"total":4600000000}
data: {"status":"success"}

System information
curl http://localhost:8100/api/system/info | jq

{
  "hostname": "gb10-louis",
  "os": "Ubuntu 24.04.2 LTS",
  "kernel": "6.8.0-49-generic",
  "arch": "aarch64",
  "cpu": "72 cores",
  "ram": "128 GB",
  "gpu": "NVIDIA Blackwell",
  ...
}

Docker Configuration

Dockerfile (gb10-manager)

FROM ubuntu:24.04

# Required system tools
RUN apt-get update && apt-get install -y --no-install-recommends \
    bash ca-certificates curl iproute2 procps \
    pciutils usbutils dmidecode lshw inxi \
    python3 python3-pip openssh-client sshpass \
    && pip3 install --break-system-packages webssh \
    && apt-get clean && rm -rf /var/lib/apt/lists/*

# Docker CLI (pas le daemon)
RUN install -m 0755 -d /etc/apt/keyrings \
    && curl -fsSL https://download.docker.com/linux/ubuntu/gpg \
       -o /etc/apt/keyrings/docker.asc \
    && echo "deb [arch=arm64 signed-by=/etc/apt/keyrings/docker.asc] \
       https://download.docker.com/linux/ubuntu noble stable" \
       > /etc/apt/sources.list.d/docker.list \
    && apt-get update \
    && apt-get install -y --no-install-recommends docker-ce-cli \
    && apt-get clean && rm -rf /var/lib/apt/lists/*

COPY gb10-manager /app/gb10-manager
RUN chmod +x /app/gb10-manager

WORKDIR /app
EXPOSE 8100
CMD ["/app/gb10-manager"]

Commande docker run

docker run -d \
  --name gb10-manager \
  --network host \
  --gpus all \
  --restart unless-stopped \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -v /etc:/host-etc:ro \
  gb10-manager:latest

Internationalization (i18n)

The interface is fully available in 5 languages. Translation is handled client-side via the I18N object in app.js and the HTML attributes data-i18n.

Supported languages

Translation mechanism

HTML - attribut data-i18n
<button data-i18n="refresh">Actualiser</button>
<span data-i18n-title="settings">Parametres</span>
<input data-i18n-placeholder="chatPlaceholder">

JavaScript - translation function
function t(key) {
    return (I18N[currentLang] && I18N[currentLang][key])
        || I18N.en[key] || key;
}

Security

Dummy / Expert Mode

The security validator (security/validator.go) analyzes each command submitted and blocks dangerous operations in Dummy mode:

• Commands rm -rf, mkfs, dd, shutdown, reboot
• Modifications de /etc/passwd, /etc/shadow
• Redirection to devices (/dev/sda, etc.)
• In Expert mode: all commands are authorized with user validation

Authentification

System operations (hostname modification, DNS, etc.) require authentication via sudo password. The password is transmitted only for the current operation and is never stored.

Secured container

• The Docker socket is mounted read/write for container management
• The host /etc files are mounted read-only (:ro)
• The container uses --network host for direct access to local services
• Le flag --gpus all provides access only to nvidia-smi, no unsupervised direct GPU access

Installer

• SSH password encrypted in memory with AES-256-GCM (PBKDF2 key derivation)
• Never written to disk in plain text
• SSH connection with 10-second timeout
• Keyboard-interactive support in addition to password authentication