Backend.AI Reservoir

Ecosystem

Reservoir

-
Package Repository

Enterprise Feature
R2

Independent Package Repository and Update Service for Backend.AI Clusters

Backend.AI Reservoir is a local package repository service designed for air-gapped environments, private clouds, and offline sites. It securely delivers essential open-source and operating system packages for AI/ML development and operations within your internal network.

If you want to reduce the security and operational burden of configuring firewall exceptions and importing packages, Backend.AI Reservoir provides a safe solution. Access the packages you need quickly through periodically synchronized repositories.

*Reservoir is available as an additional component of the Backend.AI Enterprise suite.

Air-gapped AI infrastructure

Complete AI environments in air-gapped networks

Keep AI models and tools up to date, even in air-gapped networks. Reservoir completes the software supply chain for isolated AI infrastructure.

Reservoir AI

On-premises model registry

A model hub for managing and deploying AI model weights in air-gapped environments. Synchronize HuggingFace Hub models to your internal network and manage versions in a Git-like workflow.

Offline model mirroring — Sync HuggingFace Hub to internal network
Model version control — Git-like tagging, branching, commit-based history
Fine-grained access control — Per-model view/download/deploy RBAC
Metadata management — License, benchmark data

Reservoir

Local package repository

A local repository for safely providing programming language and OS packages in air-gapped environments. Enables developers to use the latest libraries.

Python packages — PyPI air-gapped mirror, use pip install as-is
OS packages — Ubuntu, AlmaLinux system library provisioning
Security scanning — Package integrity checks and license scanning
Auto sync — Only policy-approved packages mirrored internally

Package Management in Air-Gapped Environments Requires Significant Effort.

Many enterprises and research institutions operate AI research and services in completely isolated air-gapped environments due to security policies. In these environments, AI libraries, frameworks, and datasets cannot be downloaded directly from external repositories. Organizations must acquire all necessary packages and dependencies externally, then import them into the internal network.

The package installation and update process, including security scanning, is cumbersome and time-consuming. Despite these efforts, malicious packages can still enter the network. For complex AI stacks with intricate dependency relationships, risks remain from missing specific versions or incompatible packages.