AI infrastructure for everyone

MAX is a free and permissive AI inference framework that enables developers and enterprises to develop and deploy AI inference workloads on any hardware type, into any type of environment (including into production). You can read more about MAX here.

Mojo is a new programming language that has the expressiveness of Python, with the performance of C. Mojo is a member of the Python family and powers critical parts of our infrastructure. Mojo is completely optional to use, as MAX supports Python, C++ and other languages natively. You can read more about Mojo here.

MAX Enterprise is a cloud platform we are developing for enterprises who seek managed services and enterprise support from us for AI infrastructure.

MAX and Mojo are governed by a modified Business Source License which we call the MAX & Mojo Community License. We have adopted this license to ensure that we provide maximum flexibility to the world to utilize our infrastructure while balancing our ability to commercialize our infrastructure for the long term.

‍

Modular open sources a lot of its technology, including the Mojo Standard Library, implementations of deep learning models like Llama 3 and more. This code is available on GitHub and it is licensed under Apache 2 with LLVM Exceptions, and includes contributions we receive from the community. These contributions are incorporated in MAX and Mojo and their usage and distribution (as part of MAX and Mojo) are covered by the broader MAX & Mojo Community License. We strongly believe in open-source software and are committed to continually opening our platform over time.

For MAX and Mojo, we took the modified BSL approach because it allows enterprises, developers, and our rapidly expanding community to copy, modify and redistribute their applications under a wide range of conditions, including use in production. There are many other licenses like AGPL and SSPL that impose copyleft requirements and are problematic for many use cases. We believe our licensing approach strikes the right balance for both MAX and Mojo and only restricts use for a narrow range of use cases that are competitive with Modular’s business activities.

‍

For our Github repositories and open sourced code, along with contributions from our community - we strongly believe in the Apache 2 with LLVM Exceptions, as it provides legal protection to users and contributors of the software. We outline more thoughts in this blog post.

You are free to build, extend and develop anything on top of MAX and Mojo, including products and services that you want to monetize or that you can derive an income from (i.e. content on YouTube, Twitch, etc.), and you can put them in production. You can even build products that are competitive to Modular, as long as you aren't commercializing them. We have a specific provision defined for what we deem a "Competitive Activity" that captures what you can’t do in our License.

To ensure that we can drive a commercial model for our business, we define a Competitive Activity as building a commercial, publicly available product or service whose primary purpose is to provide AI infrastructure services to third parties, or associated paid support for this purpose. 

‍

Outside of this definition, you are free to build whatever you want with MAX and Mojo and integrate it into your products or services. For example, if you are building a startup whose primary purpose is AI code generation, or search, or health agents, you can use MAX and Mojo freely as part of your startup. Similarly, if you want to educate folks on YouTube on MAX and Mojo, you can do that freely relevant to our Trademark Policy. Conversely, if you want to build a startup whose primary purpose is to provide managed artificial intelligence services to third parties in a cloud hosted platform using MAX and Mojo for profit, its primary purpose overlaps our definition, and that would be deemed a Competitive Activity, which you cannot do without written permission.

‍

We have done this to ensure existing organizations whose commercial models serve the primary purpose of providing AI infrastructure services can’t simply take our infrastructure and offer it themselves without our permission. We are building MAX Enterprise for this purpose, and as the creators, only we can provide the world's best enterprise and developer support accordingly. If you’d like to discuss licensing or commercialization of MAX and Mojo in a way that seems competitive, please contact our Sales Team – we’re very friendly.

If you want to build software that is not competitive today, but becomes competitive in the future, the License protects you – it will not be considered competitive. You will be free to continue to develop, iterate and extend your application.

MAX is free to utilize and deploy in research and production environments on your own infrastructure, as long as your use case is not a Competitive Activity.

Yes, you are free to use MAX to build any production applications you want, including deploying MAX yourself, as long as your use case is not a Competitive Activity.

Yes, just make sure your application is not a Competitive Activity and follows the terms of the License.

No limitations, you may scale MAX to any workload and size you need for supported hardware.

Yes, we aim to release MAX support for NVIDIA GPUs in fall 2024 and that will use the same approach as the current CPU support.

Yes, that is available under our MAX Enterprise version of MAX. You can learn more on our Enterprise page, or contact our Sales Team.

Yes, we are building a managed AI cloud platform for MAX Enterprise with early design partners for AI infrastructure services. Please Contact Sales if you are interested in becoming an early partner.

MAX and Mojo today are licensed under the same License as indicated above. You are free to use them to build anything you want, including in production, as long as you aren’t building something that is a Competitive Activity. Mojo is a very horizontal programming language and we’d love to see it used to build many use-cases that are not currently solved with other programming languages.

You are free to build, extend and develop anything with Mojo as long as it's not a Competitive Activity within the scope of our License.

We are open to distributing Mojo without MAX in the future, but today MAX is deeply coupled to Mojo and distributing them together as a single package allows us to move faster. We are already having trouble keeping up with community activity, and are optimizing our ability to move fast and build the world’s best platform for heterogeneous compute. We also want to learn and evolve with our community, and learn how Mojo is being utilized with MAX.

Accelerators require runtime components and orchestration logic that falls into the domain of MAX, not Mojo. The best way to use Mojo with GPUs and other accelerated computing is inside the MAX Framework. The Graph APIs and other extensibility APIs allow you to run Mojo code directly on GPUs and other hardware.

Yes. We remain strong believers in the importance of open-source software to accelerating innovation and are deeply committed to progressively make source code of Mojo fully available over time.

Modular open sources a lot of its technology, including the Mojo Standard Library, implementations of deep learning models like Llama 3 and more. This code is available on GitHub and it is licensed under Apache 2 with LLVM Exceptions, and includes contributions we receive from the community.

‍

We strongly believe in the Apache 2 with LLVM Exceptions, as it provides legal protection to users and contributors of the software, and we outline more thoughts in this blog post. All contributions are incorporated in MAX and Mojo and their usage and distribution (as part of MAX and Mojo) are covered by the broader MAX & Mojo Community License.