TL;DR: llm.codes converts JavaScript-heavy Apple docs (and 69+ other sites) into a clean llms.txt that AI agents can actually read.

Quick Start: Try it now with Apple’s Foundation Models docs: llm.codes

Even the smartest models can’t fetch fresh docs - especially when the docs are hidden behind JavaScript. While working on Vibe Meter, Claude tried to convince me that it wasn’t possible to make a proper toolbar in SwiftUI and went down to AppKit. Even when I asked it to google for a solution, nothing changed.

The Real Problem: JavaScript-Heavy Documentation

The core issue? Apple’s documentation heavily uses JavaScript, and Claude Code (or most AI agents to date) simply cannot parse that. It will fail and see nothing. So if you’re working with a component where documentation only exists on JavaScript-rendered pages, you’re completely stuck.

Enter llm.codes

That’s when I built the docs converter. llm.codes allows you to point to documentation and fetch everything as clean Markdown. While it’s optimized for Apple documentation, it supports a wide range of developer documentation sites. Here’s what you get:

• Your AI can finally see Apple docs - No blind spots from JavaScript pages
• 70% smaller files - More context space for your actual code
• Works with 69+ sites - AWS, Tailwind, PyTorch, PostgreSQL, and more

llm.codes uses Firecrawl under the hood, and I pay for the credits to keep this service free for everyone.

Real-World Example

Remember my toolbar problem? Here’s what happened: I dragged the generated SwiftUI markdown from my agent-rules repository into the terminal, and suddenly Claude wrote exactly the code I wanted.

The key insight: When you work on a component, just ask Claude to read the docs. It will load everything it needs into its context and produce vastly better code.

For people who think @Context7 is the answer: if you use the context7 mcp for SwiftUI, you get sample code from 2019, which will produce horribly outdated code. You need current documentation, not ancient examples.

Beyond Just llm.codes

I used this trick before in my post about migrating 700 tests to Swift Testing. With llm.codes, you get significantly smaller markdown files, which preserves more token context space for your agent.

I also maintain a collection of pre-converted Markdown documentation files in my agent-rules repository, that go beyond just documentation.

The converter itself? Completely vibe-coded with Claude and open source. I chose the stack (Next.js, Tailwind, Vercel) but didn’t write a single line of TypeScript-and it worked beautifully on the first try.

Try It Out

Convert a page →

No sign-up needed.

AI agents are the future of coding. Until docs catch up, llm.codes is your bridge to that future.