Mogan STEM is a TeXmacs fork aimed at formula-heavy scientific documents

A fork of TeXmacs with the emphasis moved to speed

The lineage explains the shape of the project. Mogan STEM is deeply optimized from GNU TeXmacs, with emphasis on performance and user experience, and the readme frames it as a professional scientific writing platform targeted at mathematics, physics, statistics and computer science, specially designed to handle complex formula-based documents. The claim it makes is speed, the ability to create scientific documents up to 100x faster than with LaTeX, and the readme backs that with a specific comparison, faster startup optimized compared to TeXmacs and a modern Qt6-based user interface. The reason the fork exists rather than a patch upstream is visible in the repository layout, which carries the TeXmacs tree as a subdirectory alongside src, 3rdparty, packages, tools, web, devel and a GSoC directory, and the build is driven by an xmake.lua file rather than a traditional autotools setup. The topics attached to the repository also name the technology stack honestly, since scheme, qt, cpp, wasm, emcc and xmake all appear there.

Magic Paste is the feature that comes from the AI era

The first named feature is Magic Paste, and it is the reason a fork of an older editor exists at all. It supports mainstream AI models such as ChatGPT, Claude and Gemini, and it automatically recognizes mathematical formulas, code blocks, tables and other complex formats, converting them to the standard TeXmacs format with one click and no manual adjustment required, while preserving original formatting and style integrity. The practical problem it addresses is specific. Copying a formula out of a chat window or a PDF gives you either a broken piece of text or a LaTeX string, and pasting either into a structured editor is fiddly. A converter that recognises the shape of the content and emits the right object removes that step. The interesting boundary is that the model is a choice rather than a fixed one, so the feature works with whichever provider you already pay for, and in the community edition the wiring is presumably yours to supply.

Export paths decide whether a document can be submitted

The second feature block is about getting the document out, and it is where a writing tool either succeeds or does not matter for academic work. Batch export in multiple formats is supported to suit different scenarios. PDF export preserves the original document layout and supports high-resolution printing, and LaTeX export improves the compilation success rate and is compatible with mainstream journal submission requirements. That second half is the important claim, because a journal that accepts only LaTeX effectively disqualifies any editor that cannot produce it. Batch export matters for a different reason, which is that a thesis or a paper set is many files and converting them one at a time by hand does not scale. The document formats supported list names TeXmacs, LaTeX, PDF, HTML and more, so the editor is not a closed format even though the native one is a structured document format inherited from TeXmacs.

Math editing is a mode, a shortcut system and a symbol table

The third feature is fast math editing, and the readme breaks it into three parts that each matter to someone. An independent math mode allows quick switching and focused formula editing, so the document view is not permanently rearranged around mathematics. Smart shortcuts use Lego-like memory rules and Tab-key cycling to reduce mouse operations, which is a claim about muscle memory rather than about capability. A rich symbol library covers a complete set of mathematical and physical symbols with numerous key combinations. The larger claim sits in the fourth feature, easy to learn and capable of thesis-level editing, with three primary modes that switch seamlessly, a clear structure for different content types, a well-organized document hierarchy, and click to insert environments for complex typesetting. Whether the learning curve is really that short is a question you should test on your own document rather than take from a readme, but the structure of the claim, a quick start for casual use and depth for thesis writing, is the right shape for the audience.

The plugin system is the real extension surface

Under integration with other scientific tools, the readme lists a plugin system extensible via multiple programming languages, naming Python, C++ and Scheme, alongside the document formats and interactive plotting with integration of data analysis tools for interactive plots. Three languages for plugins is an unusual and deliberate choice, and it is a direct consequence of the fork's design. Scheme is the language of the editor's own core, so an in-process extension is available to anyone who knows the internals, C++ is there for the parts that need to be fast or that need to reach native libraries, and Python is there for the people who want to plot or analyse without leaving the editor. Interactive plotting is the concrete instance, since a scientific writing tool is not finished when it can typeset a formula, it is finished when a figure generated from the data next to it is one action away. The Scheme option also means the extension surface is as deep as the original.

Community edition and commercial edition are different products

The readme describes two official websites and the difference matters for anything involving the AI features. Mogan STEM, the community version, is at mogan.app, and Liii STEM, the commercial version, is at liiistem.com, described as free with built-in AI and OCRs. So the commercial edition bundles the AI and OCR capabilities while the community build does not, which is worth understanding before you install the wrong one, especially since the Magic Paste feature is presented in the community build's own feature list. The repository also carries ai-docs at the top level, which suggests the AI integration has its own documentation set separate from the main user manual. The community version is the one published from GitHub releases, and the commercial version is installed from its own site, with the readme pointing there for installation options.

Building from source needs an xmake toolchain and a platform guide

The source route starts with a clone.

git clone https://github.com/MoganLab/mogan.git
cd mogan

After that the readme deliberately does not give you a single command, and instead points to three separate development guides, for GNU/Linux, for Windows and for macOS, each covering detailed setup, dependency installation, build and run instructions. That structure is a signal about the build. The project uses an xmake build definition, xmake.lua at the top level with a companion directory, and the repository also carries a 3rdparty directory, so the dependency graph is substantial enough that the per-platform instructions are doing real work. The repository layout shows a src directory alongside 3rdparty, packages, tools, web, devel, tests, hooks, lolly, moebius, OSPP and a bin directory, and the .clang-format file at the top tells you the C++ side is format-governed. The release list shows a steady cadence of dated tags, with v2026.3.7 on 2026-09-29 and the v2026.3.8-rc.1 candidate on 2026-10-16, so the release candidates are a public phase rather than a surprise.

Where Mogan STEM is the wrong choice

The first wrong case is a team whose entire toolchain is LaTeX, including a version control workflow where the document is a .tex file reviewed in a pull request. Mogan works the other way round, with LaTeX as an export format, and a structured document format does not diff or merge like source, so the whole collaborative editing story is different. The second wrong case is a user who needs the AI features in a self-hosted or offline environment, because in the community build the model provider is your choice and in the commercial build the OCR and AI are bundled, so the choice of edition is really the choice of how much you want to depend on someone else's service. The third is a document with a heavy custom macro requirement, because TeXmacs-style customization is a different skill from writing LaTeX packages. The fourth is a licence-sensitive organisation, since GPL-3.0 applies to the code you link against and to modifications you distribute, which is the same position as the upstream TeXmacs. The fifth is a user who just wants to write a paper and never had trouble with an editor, because the readme's own framing is that this is a fork for a particular community, and the honest test is whether the interactive document model fits how you already think.

Editorial conclusion

Writers who want a visual editor for formula-heavy documents, especially people who already know TeXmacs and want a faster and better maintained fork, are the people Mogan STEM is for, and the AI-assisted paste that turns a formula or a code block from a chat window into a correct object is the feature that distinguishes it from the ancestor. It is the wrong choice for a team that needs a LaTeX-first workflow, since the document model is a structured document format rather than a source file, and it is the wrong choice if you cannot accept GPL-3.0 for the codebase you link against. Check three things before you commit: which edition you actually want, because the community version at mogan.app and the commercial Liii STEM at liiistem.com differ in exactly the AI and OCR features that the community build leaves out, that the export path your submission target needs is covered, since the LaTeX export is described as compatible with mainstream journal requirements, and that the plugin path you need exists, since the extension system is documented as working through Python, C++ and Scheme. The licence is GPL-3.0, the newest release is v2026.3.8-rc.1 from 2026-10-16, and the last push was on 2026-10-20.

Frequently asked questions

Official sources

1. License: GPL-3.0
2. MoganLab/mogan on GitHub
3. Project website
4. README
5. Releases