Self-hosted service
Project-MONAI/MONAI avatar
Project-MONAI/MONAI

MONAI's default branch and its latest container both track unreleased code

AI Toolkit for Healthcare Imaging

8,843 stars1,722 forksPythonApache-2.0

At a glance

What is it?
MONAI is a deep learning framework for medical imaging built on PyTorch, published as a package, a container, a weekly preview and a source checkout. The readme is unusually clear that the default branch is the unreleased version and that the container's floating tag follows the same branch, which means the two obvious ways of getting the code do not give you the same thing. The packaging adds two more edges: PyTorch is a build requirement rather than only a runtime one, and a missing deep learning framework at build time produces a warning rather than an error.
Who is it for?
MONAI fits a clinical or research group that wants pre-processing, network implementations, losses and metrics for volumetric data in one place, and that wants compositional, portable APIs that fit into existing workflows. It does not fit a deployment that expects a package tag and a container tag to mean the same build, because here the container's floating tag follows the development branch while the package index does not.
Can I use it commercially?
Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
Is it still maintained?
Yes. The repository last received commits 2 days ago.
What is it written in?
Mainly Python, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • Project-MONAI/MONAI8,843 ★ · +166
Star history
Repository2026-09-112026-11-15
Project-MONAI/MONAI8.7k8.8k

166 Stars in the last 65 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

The default branch is the unreleased version

The requirements section says it plainly. The current state of the development branch in this repository is the unreleased version, which will typically support current dependency versions and include fixes and updates to get there. A separate sentence says major releases have their dependency versions stated for them. Consequence: the branch you land on by cloning is not the release you get from the package index, and the readme tells you so rather than letting you find out. That is more honesty than most projects manage. The framework support window is described separately as the current release plus three previous minor versions, with a note that a compatibility problem with other dependencies can delay support for a version until a major release.

The container's floating tag follows the development branch too

The container section says the image is tagged as latest for the latest state of the development branch, or with a release version. And the quick-start command it gives you is:

code
docker run -ti --rm --gpus all projectmonai/monai:latest /bin/bash

Consequence: two things at once. A container pulled by the floating tag and a package installed by name are different code under names that look parallel, and the quick-start command asks for a container runtime with GPU access even though the library itself runs on a CPU. So the fastest documented route to a working environment is the one that requires hardware most deployments do not have, and the tag on it is the one that moves. A second, slimmer image definition is provided for building locally, so there are two images to choose from and the readme does not say what the slimmer one leaves out.

Installing from source downloads the deep learning framework

The build requirements are not just packaging tools. They include the deep learning framework at a minimum version and the numeric library with an upper bound, alongside the build backend, a build accelerator, a version helper and a backport of a standard library module:

code
  "torch>=2.8.0",
  "numpy>=1.24,<3.0"

Consequence: installing from a checkout pulls the largest dependency in the ecosystem before anything can be built, on a machine that may not have a compatible wheel for it. Both are normal consequences of shipping compiled extensions against a fast-moving framework, and both make a source install a considerably heavier operation than the package index is. The same two libraries appear again as the runtime dependencies, with the numeric library carrying an upper bound that excludes the next major.

A missing framework at build time produces a warning

The packaging script tries to import the deep learning framework and, on any of four failure kinds, emits a warning saying the extension build was skipped and carries on. Whether it builds the native extensions at all is decided by an environment variable (BUILD_MONAI), and a second variable (FORCE_CUDA) is documented in a comment as being ignored when the first is off. Consequence: you can install the package, get no native acceleration, and have nothing but a warning to tell you. That is the right behaviour for an install on a machine with no accelerator, and it also means what you installed depends on two environment variables whose defaults are off, which is worth knowing before you benchmark. The version of the framework that is present is then folded into a single integer so the native code can be compiled against the right API, which is why the import has to succeed before anything else happens.

Three channels and a fourth combination track three cadences

The links section lists the release on the package index, the development branch, a separate weekly preview package on the same index, a conda-forge channel, a container, and a source checkout. Consequence: there are at least four distinct combinations of code available to you and nothing in the readme says which one any given tutorial assumes. The weekly preview is the one most people have never heard of. If you are following a tutorial, the fastest way to be wrong is to have installed one channel and be reading documentation written for another. The header carries badges for the package index, the container and the conda channel.

The container fetches a vendor tool and checksums it in transit

The image definition downloads a command-line archive from a vendor URL, unpacks it, computes a checksum for every file, sorts them, and verifies them against a checksum file that arrived in the same download. It then adds the directory to the path, twice, and installs a system library for slide images. Consequence: the verification proves the archive was not corrupted in transit rather than that it is the tool you intended to run, and two copies of that path entry are left in the environment. A TODO comment in the file also points at an issue in the numcodecs project about revising the Dockerfile. The base image itself is parameterised, with a build argument naming a specific vendor image and a comment telling you how to override it, so the one part of the definition you are most likely to want to change is the easiest.

Two documentation sets are published side by side

The links section offers API documentation for the milestone release and API documentation for the development branch, separately, plus a wiki and a project tracker. Consequence: a link can take you to documentation for code you do not have installed, and because the default branch is the development one, the second of those two is the set that matches a fresh clone. For a project on a rolling release that is the correct arrangement, and it is also the single most common way a reader ends up debugging a method that does not exist. The readme also points at technical highlights and a What's New page for the milestone releases, which is where the difference between what you have installed and what the documentation describes should show up first.

A C++ classifier on a pure Python package

The project metadata lists a C++ programming language classifier alongside the Python ones and the typing marker. Consequence: that is there because of the optional native extension, which is built conditionally and can be skipped. It means the distribution advertises a compiled component while a plain install may not include it, so the classifier describes the project rather than any particular install of it. Combined with the extension being gated on environment variables, it is worth checking whether you got the accelerated build before you measure anything. The metadata is deliberately sparse elsewhere too: the version is declared dynamic rather than written down, so it comes from the release process rather than from the manifest.

Editorial conclusion

MONAI fits a clinical or research group that wants pre-processing, network implementations, losses and metrics for volumetric data in one place, and that wants compositional, portable APIs that fit into existing workflows. It does not fit a deployment that expects a package tag and a container tag to mean the same build, because here the container's floating tag follows the development branch while the package index does not. It also does not fit a build environment that cannot fetch PyTorch, since PyTorch is a build requirement. Before adopting it, decide which of the four channels you are pinning, and check whether you want the native acceleration at all before you expect it.

Frequently asked questions

what is monai

A PyTorch-based open-source framework for deep learning in healthcare imaging, part of the PyTorch ecosystem. It provides flexible pre-processing for multi-dimensional medical data, compositional APIs meant to integrate into existing pipelines, domain-specific implementations of networks, losses and evaluation metrics, customisation for different levels of expertise, and multi-device data parallelism.

What is the meaning of MONAI?

Medical Open Network for AI. The readme expands the acronym that way and then states the project's ambitions: a community of academic, industrial and clinical researchers on a common foundation, end-to-end training workflows for healthcare imaging, and a standardised way for researchers to create and evaluate models.

how to install monai

One command installs the current release from the package index. Other options are pointed at an installation guide, and a container image is published as well, tagged either with the development branch's latest state or with a release version. Installing from a source checkout is a heavier operation, because the build requirements include the deep learning framework itself.

what is monai in python

It is a Python package of that name, requiring Python three point ten or newer, distributed through the package index and a conda channel with a weekly preview package alongside. Its metadata also carries a C++ classifier because of an optional native extension, which is built only when certain environment variables are set.

how to use monai

The readme points at two notebooks that run in a hosted notebook service, one a demo on medical image classification and one aimed at developers of the library itself. Longer examples and tutorials live in a separate tutorials repository, and the technical documentation is hosted separately, with different versions for the milestone release and the development branch.

is monai open source

Yes, under a version-two Apache licence held by a consortium rather than an individual, with a licence file at the repository root. The dependency support policy it publishes is narrower than the licence suggests: versions are supported for up to two years where possible, and a discovered vulnerability may cause a version to be explicitly excluded.

Official sources

  1. License: Apache-2.0
  2. Project-MONAI/MONAI on GitHub
  3. Project website
  4. README
  5. Releases
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/project-monai-monai.svg)](https://hysenlabs.com/projects/project-monai-monai)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/project-monai-monai.svg)](https://hysenlabs.com/projects/project-monai-monai)

Was this useful?