req4cmt

A comment plugin for your blog or website, just like disqus without ads.

It requires NO cookie, no signups, no login, no bullshit. Just type your comment in a textarea and submit.

To render the textarea, a few kilobyes static uncompressed req4cmt.js is loaded.

Then the .js do a fetch and render the comment list

Runs on Cloudflare worker for free. Every data is self-contained in a git repo

inspired by staticman and its successor comment-worker, transfer HTTP POST content, append to a JSON file, commit to a git. May even allowing additional blob for picture attachments in the future.

Instead of using Github proprietary API, req4cmt use isomorphic-git to speak the git-http protocol, enabling read/write to any git remote.

评论插件。通过cf worker把内容写入到git-http远端，不依赖github api

Demo

Visit my blog: https://blog.est.im/2025/stdout-07

Setup

1. fork the repo and deploy to Cloudflare Worker. Assign a domain or use the default like req4cmt.myaccount.workers.dev
2. Create another empty repo for data storage, like github.com/gh_user/my_comments
3. in your Worker settings, create a new environment secret
4. the secret name is REPO, value is the git http url like https://req4cmt:PAT@github.com/gh_user/my_comments.git
5. the PAT in the above url is a fine-grained Personal Access Token with just one repo scope to gh_user/my_comments with content read/write permission. You can create a token under your github personal settings - Developer Settings.
6. if you are using Gitlab or others, similar private tokens can be found
7. embed a snippet to your HTML page. <script defer src="https://req4cmt.myaccount.workers.dev/req4cmt.js"></script>
8. A new <div> with a <form> and a <dl> will appear for your HTML page just below the <script> tag

The UI is too ugly? Modify the <div id="req4cmt_thread"> inside dist/req4cmt.js yourself.

Plans

• git committer
• attachments
• Github pull requests for moderation
• add cache and reduce API calls
• rich content formatting like pre or markdown
• at someone for notification?
• also as a Github App
• limit length
• 20250912 git-http fetch file for private repos, replace github download.

failures and non-goals

1. partial fetch (2025-09-13)

To accelerate git clone, in theory we only need to fetch the latest commit and the individual .jsonl file.

I failed attemped to implement this. isomorphic-git does not handle single file checkout well. When .commit() Other files gets deleted. WTF?

Also the git.clone() at minimal depth=1 would still load every .jsonl from a shallow commit.

Scaling might be an issue in the future.

2. no append to git commits

How to append new lines to a large .jsonl file, without downloading the previous file content?

git use sha1 for hashing (mostly). sha1 can be "streamed" by serializing its internal state, in Python you can pickle a hashlib.sha1 instance.

Sadly, the git blob is hashed as "blob " + <decimal size> + "\0" + <file content>

every new line is appended, the <decimal size> changes, makes the header different. To make a new git commit, download the entire file, re-hash is a must .

Learned from ChatGPT today (2025-09-30). Saved me lot of time.

3. why not use SQLite/D1 or some proper database?

The same reason why choosing static site over Wordpress, which uses a database

I dislike managing DBs. Think of all those mess with backups, migrations, imports, exports, difference between mysql/pg/sqlite/etc. Tons of operating cost just for the sake of few blog comments

4. wasted CPU and memory, git is too heavy for comments

For a blog comment plugin, most of IOs are read. The worker will load single .jsonl from raw.githubusercontent.com if you are using github.

In case of writes, req4cmt worker will clone the repo (as minimal as possible), append the new line, and commit it back to the remote.

Let's be realistic, your blog won't have millions of comments. The clone should be fast.

And optimistically, There won't be too many visitor writing comments at the same time. So a basic git optimistic lock should be good enough

5. comments are risky, moderation is hard!

It's a git of .jsonl files, one comment for one line per commit, just git clone, sed them and git push

To erase the history entirely, use git cherry-pick and git push -f

Not a shell user? I don't think you'd run a static generated site anyway.

Future work:

• allow req4cmt to commit to a different branch
• use "Pull Request" to review them before merge.

6. How to fight spam?

Look at the source. You can insert you clever if-else there.

or git filter-branch the commits just in case.