Martin Donath

I just finished migrating all the open source docs from MkDocs to @zensical 🚀 @FastAPI was the last one. The iteration (build) speed makes the developer / author experience so much nicer, so much more productive.

Zensical 0.0.42 improves stability and accuracy of link validation, and fixes some bugs related to file watching, abbreviations in the table of contents, and GLightbox. Release notes: github.com/zensical/zensi…

We got some heat for warning our users that MkDocs 2.0 won't have a plugin system in Material for MkDocs. The maintainer just confirmed it again and is actively working on MkDocs 2.0 – no plugins, by design. Markdown + templating only. #discussioncomment-16917686" target="_blank" rel="nofollow noopener">github.com/mkdocs/mkdocs/…

Zensical Monthly 04/2026 AI agents don't stop when your docs have a gap. They guess – confidently, fluently, and wrongly. This month's Zensical Monthly is about what that means for technical writers, and what we're building about it. Web version: mail.zensical.org/monthly/2026/0…

Zensical 0.0.40 is out with macro support. You can now use macros inside Markdown files and even docstrings – a much-requested feature that was never possible with mkdocs-macros. Release notes: github.com/zensical/zensi…

Zensical 0.0.39 is out – a bugfix release smoothing out edges on link validation and the new lightbox integration. github.com/zensical/zensi…

Pretty happy with how this turned out! We found so many issues in our own docs when dogfooding – unresolved links and unused definitions that had been silently hiding for months. 😅

Zensical 0.0.37 adds support for installable themes – you can now package your existing customizations into a theme that can be installed via pip exactly as with MkDocs. Read the release notes: github.com/zensical/zensi…

@Legion396065537 @zensical Thanks! In fact, the issue was reported for Material for MkDocs, which has been existing for 10 years, not Zensical.

@squidfunk I faced the same issue in setting up @zensical too. Claude confidently suggested wrong config options. The trick is to always doubt yourself before doubting a product. Saves me from embarrassment every time.

This is a prime example: LLM confidently hallucinating our recommended configuration, trapping the user into thinking it's a bug. Stop submitting slop and wasting our time as maintainers! Always verify before you paste: #issuecomment-4324498173" target="_blank" rel="nofollow noopener">github.com/squidfunk/mkdo…

@DarkCisum I can't wait for the subsidizing phase of LLMS to end, which we're seeing the beginning of, with token costs going up significantly.

@DarkCisum This is why we ask for a minimal reproduction. The reporter usually at least has to go through the trouble of reading our docs again, resolving the issue in the process. Now all of that burden is shifted onto us maintainers. LLM says so, prove it wrong.

@Tugrul_Guner As a maintainer, I don't care who proposes a PR, only if it fits the vision of the project and if it's something that I can commit to maintain. In my experience, LLM PRs are like LLM-generated code – often _almost_ correct.

@squidfunk So you some people here are saying that if Karpathy creates a full LLM generated code and PR to your open source repo, you’ll ban him, am I correct?

👋 Open Source maintainers: what's your experience with issues and PRs sent to your repositories that are obviously fully LLM-generated?

@bnafOg Yes, exactly. I can ask an LLM myself, so it's essentially 2026's "let me Google that for you". Interesting times ahead!

@squidfunk The quality framing misses the harder cost: even a high-quality LLM PR has no author behind it. Push back with a question and you get silence or another LLM response. The review burden — understanding the why, not just the what — shifts entirely to you.

@janaka_a I don't agree entirely. Yes, there were sloppy issue and PRs before, but the problem with LLM-generated PRs is that they look useful and correct, but are more than often just wrong.

@squidfunk End of the day it's not about AI or not right. It's about slot vs not. People sent half baked slop before AI. Just a magnitude issue now.

Python Markdown Extension priorities – or – why you should probably never allow developers to pick their own magic numbers: What runs before or after what? 🤯

Zensical 0.0.35 just arrived. What we shipped: - 🌄 Image zoom + galleries with GLightbox – 🏗️ Rich formatting in the table of contents – 🔗 Resolution of relative links in raw HTML Release notes: github.com/zensical/zensi…

Open Source software wins in the age of AI – not despite it, but because of it. Extensibility, transparency, and community ownership matter more than ever. It's at the core of everything we're building with @Zensical. youtube.com/watch?v=G1xqTj…

@karaal_berk IMHO, both can't solve the structural and architectural limitations of MkDocs and suffer from the issues mentioned in the blog post linked above.

@squidfunk what do you think about github.com/ProperDocs/pro… and github.com/jaywhj/mkdocs-…

We updated our blog article with a detailed explanation on why we believe forking MkDocs is not a viable path forward. We borrowed parts of ZAP 007, our latest proposal on the requirements identified for Zensical's module system zensical.org/spark/proposal… #why-forking-is-impractical" target="_blank" rel="nofollow noopener">squidfunk.github.io/mkdocs-materia…