Modular Platform: MAX Framework and Mojo Language in One Repository

What the Modular Platform Is and Who It Targets

The Modular Platform is an AI development and deployment platform. The README describes it as 'a unified platform for AI development and deployment' that includes two main components: the MAX Framework, an AI inference platform, and the Mojo language, a new programming language for AI development.

Teams that need to serve large language models or run AI inference at scale are the primary audience for MAX. Developers who want to write AI systems in a new language built for performance are the audience for Mojo. The repository contains both the language toolchain and the inference platform that uses it, meaning a single clone gives access to the compiler, standard library, kernel library, and inference server together.

The project is actively open-sourcing additional components over time, as the README states: 'We're constantly open-sourcing more of the Modular Platform and you can find all of it in here.' This incremental approach means that the repository's scope grows with each release, and documentation about which directories are available under which terms changes accordingly.

Developers evaluating the platform should check both the /Mojo and /max directories to understand what is currently public. The Dockerfile and build scripts at the repository root give a clearer picture of system requirements than the README does, since the README defers most setup instructions to external documentation sites.

Repository Structure: Mojo and MAX Side by Side

The repository separates its contents into two main directory trees. The Mojo language lives under /Mojo, which contains the compiler at /Mojo, the standard library at /Mojo/stdlib, and code examples at /Mojo/examples. Developer documentation for working in the standard library is at /Mojo/docs/stdlib.

The MAX framework occupies /max. The accelerator library is at /max/kernels. The inference server is at /max/python/max/serve, described in the README as providing an OpenAI-compatible endpoint. The model pipeline definitions are at /max/python/max/pipelines, which the README calls 'Python-based graphs.' Code examples for MAX are at /max/examples. Developer documentation for MAX contributors is at /max/docs.

These two tracks share a repository but serve different audiences. Engineers deploying inference will focus on the /max directories. Language contributors and researchers will primarily work in /Mojo.

Getting Started with MAX and Mojo

The README does not include install commands directly. For MAX, it points to the MAX quickstart guide at max.modular.com/get-started. For Mojo, it points to the Mojo quickstart guide at mojolang.org/docs/manual/quickstart/.

The repository uses Bazel as its build system, which is visible in the root-level files BUILD.bazel, MODULE.bazel, REPO.bazel, and the bazelw wrapper script. The pyproject.toml at the repository root specifies Python >= 3.10 as the minimum version. There is also a .pre-commit-config.yaml at the root, indicating that contributors are expected to run pre-commit hooks before submitting changes. The repository also includes a context7.json file at the root, which is unusual and may relate to AI agent tooling documented in AGENTS.md and CLAUDE.md.

The community directory shows several support channels. Discord is at discord.gg/modular, a discussion forum is at forum.modular.com, community meeting recordings are posted to YouTube at youtube.com/@modularinc, and a Meetup group exists for local events. Community meetings are linked from the README for those who want to follow development in real time. Bug reports go to GitHub Issues at github.com/modular/modular/issues/new/choose.

What the Repository Accepts and What It Keeps Closed

The README is explicit about contribution boundaries. External contributions are accepted for the Mojo standard library, the MAX accelerator library, MAX model architectures, code examples, and Mojo documentation. One component is explicitly excluded: 'We aren't accepting contributions to the Mojo compiler yet.'

This distinction matters for teams evaluating whether to contribute upstream. Bug reports go through GitHub Issues regardless of which component they concern. The Contributing Guide is at CONTRIBUTING.md. Internal developer documentation for MAX is at /max/docs, and the corresponding documentation for the Mojo standard library is at /Mojo/docs/stdlib. Both documents are aimed at developers who will be actively working inside the codebase, not end users.

Third-party dependencies downloaded during setup, such as models from Hugging Face, carry their own licenses. The README notes that users are entirely responsible for checking and validating the licenses of those third-party components, which can include weights, datasets, and runtime libraries. This disclaimer is particularly relevant for teams in regulated industries where the provenance of model weights must be documented before deployment.

Two License Regimes: Source Code and MAX Usage

The source code in this repository is licensed under the Apache License v2.0 with LLVM Exceptions. This is a permissive open-source license that allows modification and redistribution with conditions inherited from the LLVM project. The Mojo standard library, kernel library, and model pipeline code that contributors submit are all covered by this license.

However, MAX usage and distribution are covered by a separate, non-Apache license: the Modular Community License, available at modular.com/legal/community. These two licenses govern different things. The Apache license applies to source code contributions. The Modular Community License governs how the MAX product itself may be used and distributed.

Engineers who intend to build commercial products on top of MAX or distribute it as part of a service need to read the Modular Community License directly. The Apache license that covers the code they write does not automatically extend to the MAX binary or model weights distributed by Modular. This split is an important distinction for legal review, since engineers may assume that an open-source repository with an Apache license means the entire product is available under Apache terms.

Maintenance Status, Alternatives, and Recent Releases

The last push to the repository was on 2026-09-27. The version cadence shows rapid iteration: MAX 26.6 with Mojo 1.1.0 was released on 2026-09-17, MAX 26.5 with Mojo 1.0.0 was released on 2026-08-11, and MAX 26.4 with Mojo 1.0.0b2 was released on 2026-06-18. The Mojo 1.0.0 release in August 2026 marks the language's first stable version according to its own versioning.

An alternative to consider is ONNX Runtime, which also targets AI inference across hardware targets and supports an OpenAI-compatible serving layer. ONNX Runtime is a mature project with broader hardware vendor support and a longer track record in production environments. MAX differentiates itself by combining a custom language, Mojo, with the inference engine, which means the compiler and the runtime can co-evolve. Teams that do not need Mojo for kernel authoring can use either tool for inference serving, but teams interested in writing custom accelerator kernels in Mojo will find MAX the only available target for that work.

The repository build uses Bazel, which is a substantial toolchain requirement compared to pure-Python inference stacks. Teams evaluating the project should verify that their CI and developer environments can accommodate a Bazel-based build before adopting the codebase. The presence of a .claude/ and CLAUDE.md file at the repository root indicates that the project is also designed for AI-assisted development workflows.

Editorial conclusion

Teams deploying AI inference endpoints will find the MAX inference server at /max/python/max/serve a concrete starting point, given its OpenAI-compatible API. Developers who want to contribute to the language will find the Mojo standard library open for external contributions, while the compiler is not yet accepting them. Before committing to the platform in a commercial product, read the Modular Community License at modular.com/legal/community to confirm that MAX distribution terms match the intended deployment context.

Frequently asked questions

Official sources

1. Issues
2. modular/modular on GitHub
3. Project website
4. README
5. Releases