Evidence at a glance
A Formula Gets a Link of Its Own
On October 15, 2026, Simon Willison published a LaTeX renderer. It takes LaTeX source from the fragment of a URL, the part after `#`, and renders that content. With no fragment present, the page shows an editor where users can paste source and switch to a rendered view. The tool addresses a narrow but familiar need: inspect a formula and save or share it without first creating a full document.
The interesting change is not that a web page can typeset equations. It is that the content moves out of editor state and into the link. In many online editors, a recipient has to open an application, find a document, or enter the content again. Here, the URL points directly to source that can be rendered. Saving, forwarding, and viewing are gathered into one entry point, turning the tool from a temporary input box into a lightweight content carrier.
The page also lets users move between the rendered result and source editing, and makes the rendered view bookmarkable. That interaction keeps the source as something that can be changed, rather than delivering only an image of the formula. For people sharing expressions in discussions, technical documents, or chats, the main reduction is not the act of typing LaTeX. It is the extra handoff needed to ensure that someone else sees the same content.
The Link Becomes More Than an Address
The idea was inspired by Tiny Little Office, a collection of web applications from Paul Ford and Aboard Foundary. Its tools cover word processing, outlining, presentations, spreadsheets, calculators, drawing, and trip planning. Its signature approach is to place document data in a URL fragment, making the resulting work saveable and shareable through a link. Willison borrowed not one of its editors but the underlying arrangement: let the link carry the work, and apply that model to LaTeX.
Tiny Little Office also provides an `llms.txt` file explaining how an agent can construct a URL that renders a document. The source says Claude used those instructions to generate a presentation link explaining the URL format. This illustrates how a clearly specified content format can become an interface for both people and agents. But the available material does not say that the LaTeX renderer has equivalent instructions, or establish that agents can reliably generate content in its format. Those capabilities should not be conflated.
Willison’s development prompt for Claude Code also proposed a compression path: support a prefixed form that compresses source with a custom dictionary optimized for LaTeX. That is a consequential design goal, because once content is carried in a URL, its length affects how practical it is to copy, forward, and preserve. The source gives no encoding details, before-and-after sizes, or compatibility policy. What it establishes is that compression was part of the requested design, not that its performance on long source has been demonstrated.
The Rendering Path Is Also a Security Boundary
Passing source directly to a web page for rendering means the page must handle input from links or user pastes. Willison says he asked Claude Code to look at his existing SVG-in-Markdown tool, which uses a sandboxed iframe as an isolation measure against XSS risks. The LaTeX tool also lists DOMPurify. The engineering point is practical: content preview is not only a typesetting problem. Once user input reaches the browser, the design must account for what that input can do and how the rendered result is separated from the rest of the page.
The listed dependency versions are Pako 2.1.0, MathJax 4.1.3, faster-latex 0.3.1, and DOMPurify 3.4.16. The inventory helps maintainers see which components are involved, but the source does not describe each component’s role in the call path or provide security tests or a threat model. The presence of a sandboxed iframe and a sanitization library is therefore not proof that input is safe. Anyone adopting the pattern still needs to inspect the isolation configuration, input-handling path, and dependency update process, especially the route by which external content reaches the renderer.
Lightweight Sharing Comes with Format Stewardship
The appeal of this kind of tool is that it removes a layer of infrastructure: content can be saved and forwarded as a link instead of first becoming a separate file. The source also says the LaTeX renderer borrowed URL or Gist loading from an existing SVG-in-Markdown tool. It therefore has a path for loading content from outside the fragment itself. For a personal utility, that can make it convenient to open and inspect a piece of source, but it also extends the input boundary beyond what a user has typed into the current page.
Once a link becomes a content format, maintainers have to consider whether it will remain decodable, whether its length is practical, and what happens to old links when the page changes. The available material says nothing about access controls, expiration, capacity limits, or backward compatibility, and provides no compression measurements. Technical leads can use this model for lightweight, low-risk sharing, but should define format evolution, external-input boundaries, and failure handling before putting team documents or important workflows on it. Links make a tool easier to distribute, and they distribute its maintenance obligations too.