An LLM that turns inline comments into Git diffs
Sidenote lets you comment on a rendered blog post, then an LLM writes the actual markdown diff. No forking, no pull requests.
Someone built Sidenote, a tool that lets you click on a rendered blog post, leave a comment inline, and an LLM writes the Git diff to fix it. You approve the diff, it opens a pull request. No forking the repo yourself, no editing raw markdown in a textarea. The interesting bit is the LLM step. You write “this example is wrong” next to a code block. The model reads the surrounding context, figures out what the markdown source looks like, and generates a patch. Not a suggestion in plain English. An actual diff ready to merge. This flips the usual GitHub contribution flow. Normally you fork, clone, edit locally, push, open a PR. That friction kills most casual corrections. Technical readers spot typos or outdated examples but do not bother filing issues. Sidenote bets that inline comments are low enough friction to capture those fixes. The risk is garbage diffs. If the LLM misreads context or invents facts, you get a pull request that looks plausible but breaks the article. The repo does not say which model it uses or how it validates suggestions. You still review the diff before merging, but now you are proofreading generated code instead of a human’s edit. I like the idea for documentation sites where the source is markdown and the audience is technical enough to spot bad suggestions. For prose-heavy blogs it might generate more noise than signal. The LLM cannot tell if your argument is wrong, only if your syntax is. Would I use this on my own blog? Maybe. Depends on whether the model understands MDX and frontmatter or just treats everything as generic markdown. The repo does not have examples of what the diffs look like when the LLM gets it wrong.
Source: Show HN: Sidenote – comment on your rendered blog, an LLM writes the Git diff