Writing a book in LaTeX on the web · community: general

I would like to write a book in LaTeX and have the chapters appear as web pages or blog articles. I want to be able to keep editing each chapter over and over, and update the corresponding chapter each time this happens. I want the webpages to look fairly good. I also want the final result to be printable as a single PDF file. I want all of this to be quite easy to do.

Right now I could write in LaTeX, convert some formatting commands to HTML by hand, and create either a blog article or a web page. This would not work well for references (to theorems in other chapters, for example) or citations (of previous literature).

I could probably get someone to help me do the format conversion using some automated system. @Simon Burton is already doing something like this for my Applied Category Theory Course, and that works quite well.

Ralph Sarkis (Jul 02 2025 at 11:26):

Simon Burton (Jul 02 2025 at 11:44):

I tried latexml recently, it converts latex to html.. This is what the arxiv is using to generate html pages for papers. It was looking pretty good, even though it involved a lot of fiddling with css files. But I gave up when I saw the output on commutative diagrams from tikzcd... Maybe it's worth pursuing, i found the developers on github quite helpful when i had problems..

John Baez (Jul 02 2025 at 11:46):

I'm essentially ignorant of programming, and prefer to remain as close as possible to that state of ignorance, but will accept a certain amount of learning if necessary. For the Clowder project, it starts by saying

Then it describes the "initial setup". It doesn't say if the initial setup requires that I've first do something to get the "requirements", or whether the initial setup automatically gives me the requirements. I'm afraid I need to first do something to get the "requirements". I have no idea how to install python3.6 or inkscape. So do I need to learn that?

John Baez (Jul 02 2025 at 11:47):

John Baez (Jul 02 2025 at 11:52):

I used to use something called latex2html - it allowed me to easily create a website based on my paper "The octonions". The math symbols are stored as tiny gifs, and they don't look very good. But latex2html was very easy for an idiot like me to use, and the results were quite useful, at least to me.

For some books I might write, this wouldn't matter much. I don't mind creating one or two tikz pictures "by hand" for a website version of a book, and including them as jpegs. For other books, the number of tikz diagrams would be very large.

Simon Burton (Jul 02 2025 at 12:03):

John Baez (Jul 02 2025 at 12:12):

What are the downsides of your system, the one we're using for Lectures in Applied Category Theory, @Simon Burton - apart from the fact that currently I rely on you every time I want to make a change? Presumably I could be trained to do that stuff myself?

Morgan Rogers (he/him) (Jul 02 2025 at 12:22):

(Jon Sterling left this Zulip a long time ago due to personal conflicts; he is allegedly easier to reach elsewhere.)

daniel gratzer (Jul 02 2025 at 12:35):

He's fairly active on mathstodon, where I believe @John Baez also has an account.

Simon Burton (Jul 02 2025 at 12:36):

Well we never figured out how to get latex or pdf out of it... so that would be step 1 to try and get that to work...
(A big part of this is what file format do you write in, in this case it's some kind of markdown. For latexml it would be latex... )

John Baez (Jul 02 2025 at 13:08):

Okay, I didn't notice we (= you) never managed to get latex or PDF out of it. For my book projects that will be essential.

I'd prefer to write in something close to LaTeX. I can stomach using MathJax for equations and markdown for section headers, etc., and when I'm finished writing the book converting everything to pure LaTeX.

John Baez (Jul 02 2025 at 13:08):

fosco (Jul 02 2025 at 13:37):

Pro: one can write arbitrary latex, yes, tikz; yes, quiver; yes, xymatrices; yes, whatever perversion one has can be impelemented. On the other hand, latexml has limitations

Con: it might be a tiny bit counterintuitive in the beginning, because there are idiomatic ways to place the image where you really want. Nothing that doesn't come natural after a couple days.

fosco (Jul 02 2025 at 13:38):

fosco (Jul 02 2025 at 13:39):

{% tex classes: [antex, display] %}
\begin{tikzpicture}
\draw (0,0) circle (1cm);
\draw[->] (70:1.15cm) arc (70:35:1.15cm);
\foreach[count=\j] \i in {4,3,2,1,0}{
  \node[draw, fill=white, circle, inner sep=1pt] at (\i*360/5+90+72:1cm) {\footnotesize $x_{\j}$};
  }
\begin{scope}[xshift=3cm]
\draw (0,0) circle (1cm);
\draw[->] (70:1.15cm) arc (70:35:1.15cm);
\foreach[count=\j] \i in {4,3,2,1,0}{
  \node[draw, fill=white, circle, inner sep=1pt] at (\i*360/5+90:1cm) {\footnotesize $x_{\j}$};
  }
\end{scope}
\begin{scope}[xshift=7cm]
\draw (0,0) circle (1cm);
% \draw[->] (70:1.15cm) arc (70:35:1.15cm);
\foreach[count=\j] \i in {0,1,2,3,4}{
  \node[draw, fill=white, circle, inner sep=1pt] at (\i*360/5+90:1cm) {\footnotesize $x_{\j}$};
  }
\end{scope}
\end{tikzpicture}
{% endtex %}

Ralph Sarkis (Jul 02 2025 at 13:48):

Another option that seems to be better documented is chirun. It also has a web-based frontend which may allow you to skip any coding (except what is necessary to publish the website you create).

Jonas Frey (Jul 02 2025 at 14:14):

Besides the stacks project, Lurie's Kerodon also uses Gerby. But the set-up instructions (eg here to clone the stack project) are really non-trivial.

Joe Moeller (Jul 02 2025 at 17:00):

Mike Shulman (Jul 02 2025 at 17:02):

My understanding is that pretext is not actually written in latex, but rather some kind of XML, which is kind of a dealbreaker for me. (But if I'm wrong, someone please correct me.)

Mike Shulman (Jul 02 2025 at 17:03):

It would be great if we could assemble all the suggestions in this thread into a list that summarizes the pros and cons of each, not just for John but for others who might be interested (which might include me, someday). Perhaps on an nLab page?

Joe Moeller (Jul 02 2025 at 17:06):

Emily (she/her) (Jul 02 2025 at 17:31):

For Clowder you need to separately install Conda and inkscape, but python3.6 will be installed automatically when you run the make conda-create command.

Emily (she/her) (Jul 02 2025 at 17:31):

I've tried documenting how to build it as much as I could; there's even a “GitHub action” that automatically compiles the website to here whenever a new update is pushed.

Emily (she/her) (Jul 02 2025 at 17:31):

Still, I'd advise not using Gerby if you don't want to mess too much with code. As great as Gerby is, there is a HUGE amount of things you might end up needing to do =/

Emily (she/her) (Jul 02 2025 at 17:32):

For Clowder specifically, I've essentially spent the last two months working on a fork of Gerby that would do everything I wanted it to, along with corresponding changes to the website.

Emily (she/her) (Jul 02 2025 at 17:32):

It's also relatively easy to break the project if you don't compile it often, and then if that happens you'll have no idea what you did that got Gerby to not compile...

Emily (she/her) (Jul 02 2025 at 17:32):

Maybe using plain plasTeX could be a better solution. Gerby is a fork of it, but it's not well-maintained, so it's essentially something like 2 years behind plasTeX.

Emily (she/her) (Jul 02 2025 at 17:33):

I think Forester might be another nicer alternative, I think Tim used it for translations here

Nathanael Arkor (Jul 02 2025 at 17:38):

Jonas Frey (Jul 02 2025 at 18:18):

Regarding non-latex alternatives: in the long run, I'm very hopeful about typst. This is a LaTeX replacement with an HTML export feature under active development.

Patrick Nicodemus (Jul 03 2025 at 01:15):

Patrick Nicodemus (Jul 03 2025 at 01:18):

Patrick Nicodemus (Jul 03 2025 at 01:19):

the first thing that jumps out is the rich mouseover context, the previews of related pages

Ulrik Buchholtz (Jul 03 2025 at 09:12):

Eventually we'll be able to produce accessible tagged PDFs, from which web pages can be derived via something like ngPDF. This depends on the progress of the LaTeX tagging project; something that's also required as of this month for publishing ebooks in the EU/UK under the European Accessibility Act.

Rich Hilliard (Jul 03 2025 at 13:40):

In case no one has mentioned this one: I've just started experimenting with a LaTeX package lwarp for a book-sized project. From the link:

"This package converts LATEX to HTML by using LATEX to process the user’s document and generate HTML tags. External utility programs are only used for the final conversion of text and images. Math may be represented by SVG files or MathJax. Hundreds of LATEX packages are supported, and their load order is automatically verified."

John Baez (Jul 03 2025 at 14:04):

Rich Hilliard (Jul 03 2025 at 14:51):

It was simple to install. Seems to do fine on the usual stuff: text formatting, document organization, references and linking.
It supports various approaches to rendering math and diagrams. I haven't experimented with those enough yet to choose an approach.

Kevin Carlson (Jul 03 2025 at 17:10):

Forester is not LaTeX but it allows you to write arbitrary LaTeX, critically including general tikz or whatever other library-dependent code you want, in almost the way you’re used to. We’re using it a lot at Topos recently and it has a lot of great features including exporting to the Web by default and to ordinary LaTeX in a mostly automatic way, plus very flexible linking capabilities for writing in a modular way. I suspect it’s still “alpha” enough that John wouldn’t like it but I don’t think there’s anything clearly better yet.

Kevin Carlson (Jul 03 2025 at 17:10):

@Owen Lynch you’ve thought hard about this, do you want to drop your opinions about the various options mentioned here and why you like Forester best?

Nathanael Arkor (Jul 03 2025 at 19:54):

By "use arbitrary LaTeX", you mean I can use packages like tabularray, etc. freely?

Kevin Carlson (Jul 03 2025 at 20:29):

Screenshot 2025-07-03 at 1.24.00 PM.png
Yes, you can. Here's an excerpt from a note of David Jaz's in the real Topos forest where he typesets a type theory using mathpartir; I added an example from the tabularray docs to see that Forester has no issues with that library either. It just runs your real LaTeX installation on the requisite parts of your forester document, then inserts the result into the final .xml output as SVG (I don't know any details about this processing pipeline.)

Kevin Carlson (Jul 03 2025 at 20:30):

The code looks something like this. It's a pretty analogous syntax to LaTeX, except for the somewhat annoying hygiene requirement to tag everything, including a paragraph of plain text, inside some backslash-and-braces macro.

\p{In this sketch, I will describe a simple type theory whose contexts are in one-to-one corresondence with [direct categories](djm-002L). I don't think it quite works as stated.}

\p{The rules are as follows: }

\tex{\usepackage{mathpartir}}{
\begin{mathpar}
\inferrule{\phantom{\bullet}}{\bullet \jctx }
\and \inferrule{\Gamma \yields A \jtype}{\Gamma, .x : A \yields x : A}
\and \inferrule{\Gamma \yields \mathcal{J}}{\Gamma, .x : A \yields \mathcal{J}} \\
\inferrule{\phantom{\bullet}}{\bullet : \Gamma \to \bullet} \and
\inferrule{\sigma : \Gamma \to \Delta \and \Delta \yields A \jtype \and \Gamma \yields t : A[\sigma]}{\sigma, x \gets t : \Gamma \to \Delta, .x : A} \\
\inferrule{\Gamma \yields A \jtype}{\Gamma, .x : A \jctx} \and
\inferrule{\sigma : \Gamma \to \Delta}{\Gamma \yields \Cone(\sigma) \jtype}
\end{mathpar}
}

\p{Here is a test of the tabulararray #{\LaTeX} package:}
\tex{\usepackage{tabularray}}{
\begin{tblr}{lccr}
\hline
Alpha & Beta & Gamma & Delta \\
\hline
Epsilon & Zeta & Eta & Theta \\

\hline
Iota & Kappa & Lambda & Mu \\

\hline

\end{tblr}
}

Kevin Carlson (Jul 03 2025 at 20:31):

Jon's motivation in developing Forester was exactly to allow people to "build their own Stacks project in ten minutes", incidentally, and it's true that you can probably get an example running without needing to do anything that "feels like programming."

Owen Lynch (Jul 03 2025 at 20:33):

I will say that there are like 5 different "latex to html" "compilers" and each of them works for some undefined subset of LaTeX, and I think none of those subsets include tikz; I think the only reasonable way to get good quality typesetting of figures is just to shell out to LaTeX and Forester makes this especially easy. Forester has its quirks, but I think there isn't really a serious contender in its domain.

Mike Shulman (Jul 03 2025 at 20:57):

How many of these options have a way to insert additional code into the HTML output that isn't LaTeX? E.g. to embed an interactive javascript activity or exercise?

Owen Lynch (Jul 03 2025 at 21:24):

Owen Lynch (Jul 03 2025 at 21:27):

John Baez (Jul 04 2025 at 09:54):

Thanks a million for the offer, @Owen Lynch. I'm not quite ready to dive in, which will happen when I start writing my book on combinatorics and categories. A couple of questions:

Owen Lynch (Jul 04 2025 at 09:58):

Owen Lynch (Jul 04 2025 at 10:15):

This process is slightly involved, but also extremely customizable (e.g., it is more or less straightforward to add new environments, change the document class, translate html tables into latex tables, etc.)

Owen Lynch (Jul 04 2025 at 10:18):

It looks like lwarp is more full-featured than the other latex-to-html pipelines I've seen in the past though!

Owen Lynch (Jul 04 2025 at 10:18):

Owen Lynch (Jul 04 2025 at 10:22):

Rich Hilliard (Jul 04 2025 at 11:09):

@Mike Shulman lwarp defines LaTeX environments warpprint and warpHTML to direct specific content only to the generated print or web versions, respectively.
With warpHTML you could embed javascript in the HTML with <script>.

Kevin Carlson (Jul 04 2025 at 17:48):

There are also some 595 ancillary packages, as you can see in the documentation PDF, which is a bit longer than a Lurie book; it seems like you have one for making lwarp work with pretty much every TeX library (including tikz!) This might be easier for you to get started with, John, although it’s really amazingly enormous and I wonder whether the user usually gets to avoid all that complexity.

John Baez (Jul 04 2025 at 18:11):

Since I couldn't get myself to read Lurie's books, there's no way I'll engage with a comparably large book just to learn a piece of software.

Mike Shulman (Jul 04 2025 at 20:00):

Surely that's just a reference manual and you wouldn't have to actually read it all to use the software?

John Baez (Jul 04 2025 at 20:23):

Simon Burton (Jul 07 2025 at 14:09):

Apparently the html version of the Picturing Quantum Software book is made using make4ht ( another latex -> html converter ...)

John Baez (Jul 07 2025 at 14:13):

(Hmm, that sounds like a very cowboy-ish expression, I wonder if non-Americans will understand it.)

Mike Shulman (Dec 17 2025 at 18:17):

I've started experimenting with PreTeXt, and it's actually fairly nice. The mathematics is straight-up LaTeX. The XML is only used for document structure such as sectioning, blocks, and lists; this is where LaTeX tends to use begin/end environments anyway, so the verbosity of XML is not that much greater, and appropriate editor tooling can automatically insert the ending tags for you.

The PreTeXt schema introduces a bit more verbosity, e.g. a theorem requires a <theorem> tag containing a <statement> tag containing a <p> tag before you can start stating the theorem, but that should also be automatable with a good editor (Emacs doesn't yet have such automation, as far as I can see, but I might do something about that if I continue using PreTeXt). And the reason for that verbosity is logical, since a <theorem> tag can also contain a <title> and a <proof>, which are then associated to the theorem by the XML structure and can be easily programmatically manipulated.

The XML-based design essentially forces you to separate semantics from presentation. There is no way to say within the schema "make this text 15 points tall and italics", only "here begins a section with this title". So it's not a replacement for (La)TeX as an engine for layout of printable documents. But all those things you can do with TeX are exactly the things that make it so hard to compile to other formats like HTML, so this is arguably a good thing.

PreTeXt produces HTML output that I think looks overall excellent, especially for textbooks (here are a couple examples), and the available semantic information can at least in principle improve accessibility to screen readers and so on. (And the PreTeXt people are thinking hard about this sort of thing -- for instance, you are instructed to put ending punctuation outside a display-math environment, because it is semantically not part of the mathematics and thus should be treated differently by non-visual output formats; but in visual output formats like PDF and HTML such punctuation is automatically moved to the end of the displayed mathematics, where it looks good, rather than at the beginning of the following line.) It can also be compiled to PDF by way of LaTeX, so the output looks like LaTeX since it is, and also to epub and kindle formats.

Personally, I prefer the "linear" presentation of a PreTeXt HTML document to the forester/stacks "atomized" approach, especially for something like a book that is "projected on a totally ordered way on the page axis" in the words of Serge Lang. For authoring purposes, you're free to split up your source code into multiple files as finely as you like to facilitate reorganization, using <xi:include> tags to stitch it together. Of course YMMV, but at the moment I'm happy enough with PreTeXt, compared to the alternatives I've seen, that I'm about to embark on a significant project using it.

Kevin Carlson (Dec 17 2025 at 18:43):

Tim Hosgood (Dec 17 2025 at 18:50):

it's written in markdown but exports to pure LaTeX (which is used to automatically build those PDFs whenever we update the markdown files)

(sorry, haven't caught up with this thread yet, but just wanted to throw this out as another option if it hasn't already been mentioned)

John Baez (Dec 19 2025 at 10:18):

Thanks, Tim. I don't feel anything (yet). But I will someday loop back to this topic and try to find the system that's the easiest for me to use.

Valeria de Paiva (Jan 10 2026 at 00:39):

@Mike Shulman you forgot to mention one of the main advantages of preText for me: that it can output Braille for people who have difficulties seeing. I expect that means that other alphabets are also easily covered. I wanted to use it, as many of the open source books recommended by AIM use it, but I lost my patience. glad you haven't.

Mike Shulman (Jan 10 2026 at 01:13):

I did mention "accessibility to screen readers and so on", but true I didn't mention Braille specifically.

Stream: community: general

Topic: Writing a book in LaTeX on the web

John Baez (Jul 02 2025 at 11:18):

Ralph Sarkis (Jul 02 2025 at 11:26):

Simon Burton (Jul 02 2025 at 11:44):

John Baez (Jul 02 2025 at 11:46):

John Baez (Jul 02 2025 at 11:47):

John Baez (Jul 02 2025 at 11:52):

Simon Burton (Jul 02 2025 at 12:03):

John Baez (Jul 02 2025 at 12:12):

Morgan Rogers (he/him) (Jul 02 2025 at 12:22):

daniel gratzer (Jul 02 2025 at 12:35):

Simon Burton (Jul 02 2025 at 12:36):

John Baez (Jul 02 2025 at 13:08):

John Baez (Jul 02 2025 at 13:08):

fosco (Jul 02 2025 at 13:37):

fosco (Jul 02 2025 at 13:38):

fosco (Jul 02 2025 at 13:39):

Ralph Sarkis (Jul 02 2025 at 13:48):

Jonas Frey (Jul 02 2025 at 14:14):

Joe Moeller (Jul 02 2025 at 17:00):

Mike Shulman (Jul 02 2025 at 17:02):

Mike Shulman (Jul 02 2025 at 17:03):

Joe Moeller (Jul 02 2025 at 17:06):

Emily (she/her) (Jul 02 2025 at 17:31):

Emily (she/her) (Jul 02 2025 at 17:31):

Emily (she/her) (Jul 02 2025 at 17:31):

Emily (she/her) (Jul 02 2025 at 17:32):

Emily (she/her) (Jul 02 2025 at 17:32):

Emily (she/her) (Jul 02 2025 at 17:32):

Emily (she/her) (Jul 02 2025 at 17:33):

Nathanael Arkor (Jul 02 2025 at 17:38):

Jonas Frey (Jul 02 2025 at 18:18):

Patrick Nicodemus (Jul 03 2025 at 01:15):

Patrick Nicodemus (Jul 03 2025 at 01:18):

Patrick Nicodemus (Jul 03 2025 at 01:19):

Ulrik Buchholtz (Jul 03 2025 at 09:12):

Rich Hilliard (Jul 03 2025 at 13:40):

John Baez (Jul 03 2025 at 14:04):

Rich Hilliard (Jul 03 2025 at 14:51):

Kevin Carlson (Jul 03 2025 at 17:10):

Kevin Carlson (Jul 03 2025 at 17:10):

Nathanael Arkor (Jul 03 2025 at 19:54):

Kevin Carlson (Jul 03 2025 at 20:29):

Kevin Carlson (Jul 03 2025 at 20:30):

Kevin Carlson (Jul 03 2025 at 20:31):

Owen Lynch (Jul 03 2025 at 20:33):

Mike Shulman (Jul 03 2025 at 20:57):

Owen Lynch (Jul 03 2025 at 21:24):

Owen Lynch (Jul 03 2025 at 21:27):

John Baez (Jul 04 2025 at 09:54):

Owen Lynch (Jul 04 2025 at 09:58):

Owen Lynch (Jul 04 2025 at 10:15):

Owen Lynch (Jul 04 2025 at 10:18):

Owen Lynch (Jul 04 2025 at 10:18):

Owen Lynch (Jul 04 2025 at 10:22):

Rich Hilliard (Jul 04 2025 at 11:09):

Kevin Carlson (Jul 04 2025 at 17:48):

John Baez (Jul 04 2025 at 18:11):

Mike Shulman (Jul 04 2025 at 20:00):

John Baez (Jul 04 2025 at 20:23):

Simon Burton (Jul 07 2025 at 14:09):

John Baez (Jul 07 2025 at 14:13):

Mike Shulman (Dec 17 2025 at 18:17):

Kevin Carlson (Dec 17 2025 at 18:43):

Tim Hosgood (Dec 17 2025 at 18:50):

John Baez (Dec 19 2025 at 10:18):

Valeria de Paiva (Jan 10 2026 at 00:39):

Mike Shulman (Jan 10 2026 at 01:13):