Backend.AI is a streamlined, container-based computing cluster platform that hosts popular computing/ML frameworks and diverse programming languages, with pluggable heterogeneous accelerator support including CUDA GPU, ROCm GPU, Rebellions, FuriosaAI, HyperAccel, Google TPU, Graphcore IPU and other NPUs.

It allocates and isolates the underlying computing resources for multi-tenant computation sessions on-demand or in batches with customizable job schedulers with its own orchestrator named "Sokovan".

All its functions are exposed as REST and GraphQL APIs.

This repository contains all open-source server-side components and the client SDK for Python as a reference implementation of API clients.

• src/ai/backend/: Source codes
  • manager/: Manager as the cluster control-plane
  • manager/api: Manager API handlers
  • account_manager/: Unified user profile and SSO management
  • agent/: Agent as per-node controller
  • agent/docker/: Agent's Docker backend
  • agent/k8s/: Agent's Kubernetes backend
  • agent/dummy/: Agent's dummy backend
  • kernel/: Agent's kernel runner counterpart
  • runner/: Agent's in-kernel prebuilt binaries
  • helpers/: Agent's in-kernel helper package
  • common/: Shared utilities
  • client/: Client SDK
  • cli/: Unified CLI for all components
  • install/: SCIE-based TUI installer
  • storage/: Storage proxy for offloading storage operations
  • storage/api: Storage proxy's manager-facing and client-facing APIs
  • appproxy/: App proxy for accessing container apps from outside
  • appproxy/coordinator: App proxy coordinator who provisions routing circuits
  • appproxy/worker: App proxy worker who forwards the traffic
  • web/: Web UI server
    • static/: Backend.AI WebUI release artifacts
  • logging/: Logging subsystem
  • plugin/: Plugin subsystem
  • test/: Integration test suite
  • testutils/: Shared utilities used by unit tests
  • meta/: Legacy meta package
  • accelerator/: Intrinsic accelerator plugins
• docs/: Unified documentation
• tests/
  • manager/, agent/, ...: Per-component unit tests
• configs/
  • manager/, agent/, ...: Per-component sample configurations
• docker/: Dockerfiles for auxiliary containers
• fixtures/
  • manager/, ...: Per-component fixtures for development setup and tests
• plugins/: A directory to place plugins such as accelerators, monitors, etc.
• scripts/: Scripts to assist development workflows
  • install-dev.sh: The single-node development setup script from the working copy
• stubs/: Type annotation stub packages written by us
• tools/: A directory to host Pants-related tooling
• dist/: A directory to put build artifacts (.whl files) and Pants-exported virtualenvs
• changes/: News fragments for towncrier
• pants.toml: The Pants configuration
• pyproject.toml: Tooling configuration (towncrier, pytest, mypy)
• BUILD: The root build config file
• **/BUILD: Per-directory build config files
• BUILD_ROOT: An indicator to mark the build root directory for Pants
• CLAUDE.md: The steering guide for agent-assisted development
• requirements.txt: The unified requirements file
• *.lock, tools/*.lock: The dependency lock files
• docker-compose.*.yml: Per-version recommended halfstack container configs
• README.md: This file
• MIGRATION.md: The migration guide for updating between major releases
• VERSION: The unified version declaration

Server-side components are licensed under LGPLv3 to promote non-proprietary open innovation in the open-source community while other shared libraries and client SDKs are distributed under the MIT license.

There is no obligation to open your service/system codes if you just run the server-side components as-is (e.g., just run as daemons or import the components without modification in your codes). Please contact us (contact-at-lablup-com) for commercial consulting and more licensing details/options about individual use-cases.

Run scripts/install-dev.sh after cloning this repository.

This script checks availability of all required dependencies such as Docker and bootstrap a development setup. Note that it requires sudo and a modern Python installed in the host system based on Linux (Debian/RHEL-likes) or macOS.

Please consult our documentation for community-supported materials. Contact the sales team (contact@lablup.com) for professional paid support and deployment options.

Backend.AI provides websocket tunneling into individual computation sessions (containers), so that users can use their browsers and client CLI to access in-container applications directly in a secure way.

• Jupyter: data scientists' favorite tool
  • Most container images have intrinsic Jupyter and JupyterLab support.
• Web-based terminal
  • All container sessions have intrinsic ttyd support.
• SSH
  • All container sessions have intrinsic SSH/SFTP/SCP support with auto-generated per-user SSH keypair. PyCharm and other IDEs can use on-demand sessions using SSH remote interpreters.
• VSCode
  • Most container sessions have intrinsic web-based VSCode support.

Backend.AI provides an abstraction layer on top of existing network-based storages (e.g., NFS/SMB), called vfolders (virtual folders). Each vfolder works like a cloud storage that can be mounted into any computation sessions and shared between users and user groups with differentiated privileges.

It routes external API requests from front-end services to individual agents. It also monitors and scales the cluster of multiple agents (a few tens to hundreds).

• src/ai/backend/manager
  • README
  • Legacy per-pkg repo: https://github.com/lablup/backend.ai-manager
  • Available plugin interfaces
    • backendai_scheduler_v10
    • backendai_agentselector_v10
    • backendai_hook_v20
    • backendai_webapp_v20
    • backendai_monitor_stats_v10
    • backendai_monitor_error_v10

It manages individual server instances and launches/destroys Docker containers where REPL daemons (kernels) run. Each agent on a new EC2 instance self-registers itself to the instance registry via heartbeats.

• src/ai/backend/agent
  • README
  • Legacy per-pkg repo: https://github.com/lablup/backend.ai-agent
  • Available plugin interfaces
    • backendai_accelerator_v21
    • backendai_monitor_stats_v10
    • backendai_monitor_error_v10

It provides a unified abstraction over multiple different network storage devices with vendor-specific enhancements such as real-time performance metrics and filesystem operation acceleration APIs.

• src/ai/backend/storage
  • README
  • Legacy per-pkg repo: https://github.com/lablup/backend.ai-storage-proxy

It hosts the SPA (single-page application) packaged from our web UI codebase for end-users and basic administration tasks.

• src/ai/backend/web
  • README
  • Legacy per-pkg repo: https://github.com/lablup/backend.ai-webserver

Synchronizing the static Backend.AI WebUI version:

$ scripts/download-webui-release.sh <target version to download>

Computing environment recipes (Dockerfile) to build the container images to execute on top of the Backend.AI platform.

• https://github.com/lablup/backend.ai-kernels

A programmable sandbox implemented using ptrace-based system call filtering written in Rust.

• https://github.com/lablup/backend.ai-jail

A set of libc overrides for resource control and web-based interactive stdin (paired with agents).

• https://github.com/lablup/backend.ai-hook

We offer client SDKs in popular programming languages. These SDKs are freely available with MIT License to ease integration with both commercial and non-commercial software products and services.

• Python (provides the command-line interface)
  • pip install backend.ai-client
  • https://github.com/lablup/backend.ai/tree/main/src/ai/backend/client
• Java
  • Currently only available via GitHub releases
  • https://github.com/lablup/backend.ai-client-java
• Javascript
  • npm install backend.ai-client
  • https://github.com/lablup/backend.ai-client-js
• PHP (under preparation)
  • composer require lablup/backend.ai-client
  • https://github.com/lablup/backend.ai-client-php

Plugins are detected and loaded via Python package entrypoints. The followings are the mainly used entrypoint names:

• backendai_accelerator_v21
  • ai.backend.accelerator.cuda: CUDA accelerator plugin
  • ai.backend.accelerator.cuda (mock): CUDA mockup plugin
    • This emulates the presence of CUDA devices without actual CUDA devices, so that developers can work on CUDA integration without real GPUs.
  • ai.backend.accelerator.rocm: ROCm accelerator plugin
  • More available in the enterprise edition!
• backendai_monitor_stats_v10
  • ai.backend.monitor.stats
    • Statistics collector based on the Datadog API
• backendai_monitor_error_v10
  • ai.backend.monitor.error
    • Exception collector based on the Sentry API

These components still exist but are no longer actively maintained.

The front-end support libraries to handle multi-media outputs (e.g., SVG plots, animated vector graphics)

• The Python package (lablup) is installed inside kernel containers.
• To interpret and display media generated by the Python package, you need to load the Javascript part in the front-end.
• https://github.com/lablup/backend.ai-media

• Visual Studio Code Extension
  • Search “Live Code Runner” among VSCode extensions.
  • https://github.com/lablup/vscode-live-code-runner
• Atom Editor plugin
  • Search “Live Code Runner” among Atom plugins.
  • https://github.com/lablup/atom-live-code-runner

We now recommend using in-kernel applications such as Jupyter Lab, Visual Studio Code Server, or native SSH connection to kernels via our client SDK or desktop apps.

Backend.AI supports building two types of packages:

1. Python wheels (.whl)
2. SCIE (Self-Contained Installable Executables)

To build:

# Build wheels or SCIE packages
./scripts/build-wheels.sh
./scripts/build-scies.sh

All built packages will be placed in the dist/ directory.

Refer to LICENSE file.