The Svelte docs are now available as JSON at
https://svelte.dev/content.json
But when I try to index them in Cursor, it always fails. Maybe it’s too long?
Ideas? Does Cursor need to chunk it into smaller portions maybe?
Hey, JSON format for documentation isn’t supported. You can try adding this address:
If they want to make LLM-friendly docs, what format is suggested? Markdown, I assume?
Any normal web-based format will work for the documentation scraping, as it goes through a bit of a conversion step before the AI gets provided with it.
This is actually the reason why a JSON formatted file wouldn’t work, because the scraper is expecting a normal human-readable website.
To ingest a JSON file into Cursor, it may actually be better to download the file into your codebase and then @ it into the AI as context.
@danperks Do you know if the scraper cares how large the file is? I.e. assume it’s markdown, does it need to be below a context window size, or is the scraper smart enough that it can consume any size text and process it in chunks?
Asking so I can give the Svelte team guidance, regarding whether a single large markdown is workable or would be problematic.
I found some other LLM-friendly svelte docs in TXT format
See list at: https://svelte-llm.khromov.se/
Add: https://svelte-llm.khromov.se/svelte-complete
Splitting docs into pages actually works better as it allows us to provide only the most relevant pages to the AI when answering questions. But it should work with almost anything you throw at it to some degree!