davidpapermill

Author here - this is something I've been thinking about while using Claude.ai and Claude Code. I'm often using them to prepare large documents and the UX isn't there. Thought this analogy works quite well to pin down the feeling of cognitive load.

I do agree that free fonts are overused and a really great font can work wonders. We're a little limited by licence restrictions at Papermill, so it may depend on your use case.

Love the comment on the book - I've spent too long looking at kerning and now I can only buy books that are well typeset!

Footnotes are very hard to typeset. They’re up there with tables as the biggest headache for any typesetter.

Thanks Jonny. What were you rendering?

Great! I’ll be posting videos on social media of MCP use over the next few weeks. Be sure to follow @papermillapp and keep an eye on our blog.

Worth adding that we also support the integration of many PDFs into a large document, including complex processing, but this is not part of the publicly accessible Press language (yet).

Papermill is more the "infrastructure" layer that sits below legal AI tools - we don't work directly with law firms, rather more likely to be used by a startup building a legal tool. Companies like Wordsmith, Legatics, Legora, Harvey can use Papermill to output high-quality PDFs.

That said, an export from Word to Papermill templates might be a nice addition...

Thanks Tom, good to see you on HN! You can connect Papermill to your favourite agent via MCP - we prefer Claude Code.

https://docs.papermill.io/mcp/#using-the-papermill-mcp-serve...

You can ask Claude to generate templates, full documents, ask it about the Press language, save templates, ask it about templates and more. There's a lot more to it - design guidelines for print vs web, recipes in Press etc.

I've tried it with Fable already and it's a noticeable improvement - we support visual feedback through MCP and I think that's helping Fable a lot.

Press is designed from the ground-up for agentic workflows. We spoke to hundreds of companies generating PDFs before we designed the language, and had been building AI design systems ourselves for a few years.

You can connect an agent to our MCP server, see: https://docs.papermill.io/mcp/#using-the-papermill-mcp-serve...

It's quite amazing to see what Claude can do when it has a tool like Papermill at its disposal - worth connecting and having a play.

In practice, you can either directly connect an LLM to the MCP server, or send via the API after (say) cleaning up LLM output, combining it with RAG and other data etc.

Thanks Tim! Great to see Byteful doing so well too!

Hi Folks - I'm David, CEO of Papermill. We're super excited to get feedback on what we're building. This is the first time we've posted to a forum. Designing a document language for AI is a major undertaking and we'd love to hear from folks with experience and interest in document languages and generation.

This is interesting. So it's HTML -> PDF without using Chrome/Puppeteer etc.? Is that correct. It reminds me of https://weasyprint.org/

Weasyprint (by their own admission) have really struggled with a lot of bugs, which isn't surprising if you're allowing arbitrary HTML input - how are you getting on with that?

What did you use for PDF Report Generation?

This link works: https://imgur.com/a/YtFSodj

I think the GitHub link was supposed to be: https://github.com/WellApp-ai/Well/tree/main/ai-receipt-gene...

I'm interested in this if the OP wants to update - my company has a PDF generation API, so I'm interested in seeing LLMs do this directly.

Thanks for the feedback. Yep - we've yet to update the website. It's a waitlist, we're not launched yet. Absolutely fair criticism. It'll be updated in the next few weeks!

"Very little" in terms of the text formatting I guess. But there's no reason they can't improve on that.

InDesign is awesome. Anyone designing a general purpose document language should spend a lot of time in InDesign first. LaTeX should not be your mental model.

Agreed, it's about _imperative_ vs _declarative_ languages.

The problem with languages like LaTeX etc. is they treat a document as a linear sequence of instructions: "write this text", "switch to justify right" etc. - I think that made sense when LaTeX was created, but publishing has come a long way and imperative thinking doesn't help.

A declarative form of document is the way forward. HTML is -kind of- like that, but unfortunately the assumption that a document is a linear list of stuff is still in there. That's one reason it doesn't work well for print.