Open-source Chrome extension
Comment sections, minus the junk
Bot spam, scams, abuse and sexual replies are cleaned up on X, YouTube, Reddit, Xiaohongshu and Hacker News — and each one says why. Foreign-language posts and comments are translated in place.
- Open source · MIT
- Collects no data
- 19 languages
- Pay as you go, about $1/month
Also installs from the Chrome Web Store in Edge, Arc, Brave and other Chromium browsers.
Cleanup
Junk replies, cleaned up
Bot spam, lead generation, scam testimonials, abuse and sexual replies become one line that says why. Open it whenever you like, and mark it when it got one wrong.
Every hide says why
“3 accounts posted the same line”, “mentions an account that isn't in this thread”, “looks like a scam”.
“nice” and “+1” in one line
Content-free replies move to one line at the top of the comments.
Runs become one line
Consecutive hidden replies merge into one; a thread that's mostly junk is tucked behind one line.
Case by case
See the whole page
Three accounts posting the same “trading mentor” line become one notice; the reply pulling people to another account and the abusive one say why; “nice” and “+1” move to one line.
Under a gadget post: sexual bots (“my private pics are on my profile”, “lonely guys near me”) and an airdrop bot, cleaned up with the reason.
Lead-generation comments (“check my profile”, “add me on WeChat”) are cleaned up and the same line from three accounts becomes one notice.
Fake “mentor” testimonials are cleaned up and “🔥🔥🔥” is tucked away.
The cheap-flights ad under a Hokkaido trip post is cleaned up and “nice” is tucked away.
The pages and comments are made up; the extension is real.
Mark one
Mark one spam reply, the rest of its kind go too
When ClearFlow misses one, click the funnel next to it. Replies of the same kind — reworded, from other accounts, pushing the same scheme — are cleaned up from then on.
- 1
Click the funnel
Hover a comment: a funnel appears at the end of the author line.
- 2
Its kind goes too
Reworded copies from other accounts are recognized and labelled “Same kind as spam you marked”.
- 3
Undo anytime
An Undo notice pops up; everything you marked is listed in Settings, where you can remove entries.
Under a Reddit personal-finance post, three replies push the same “trading mentor” in different words. Mark the first and the other two go too.
Under a RedNote room-makeover post, three replies pull people into the same “deals group” in different words. Mark the first and the other two go too.
The pages and comments are made up; the extension is real.
Translation
Foreign comments, read in your language
Posts and comments are translated where they are, links and formatting intact. One click shows the original.
In place
No pop-ups, no new tab: the translation sits where the text was, with the language it came from.
The original one click away
Click “Show original” to switch back, and again to return.
On X, only when it's better
X's own Grok translation is replaced only when ClearFlow's is clearly better.
Language by language
See the whole page
Under a Japanese post, comments in German, Spanish, Chinese, French and Portuguese are translated in place.
Under a Korean cooking video, comments in Portuguese, Hindi, Korean, Arabic and Japanese are translated.
A whole RedNote note — title, text and comments — read in English.
The pages and comments are made up; the extension is real.
See what was cleaned, at a glance
Click the ClearFlow icon to see how many replies were cleaned and translated on this page, today and over the last 7 days. Settings are per site: how strict each category is, and whether to translate.
- Per-site on/off
- Cleaning and translation separately
- Relaxed / standard / strict
- What was cleaned on this page
How it works
How it decides
- 1
Local rules first
Referral links, hidden contact details, the same line from several accounts, invisible characters… The rules look at structure, never keyword lists, so they work across topics and languages. No key needed, no cost.
- 2
AI for the rest
Replies the rules can't decide are sent, with the post they answer, to TypeSafe's Jev model, which answers a few yes/no questions: Is it recruiting? Abusive? Sexual? Only replies over the line are hidden.
- 3
Translation via OpenRouter
Text to translate goes to the model you pick on OpenRouter (an OpenAI model by default, DeepSeek as fallback). Only the text — no author, no post ID.
Cost
Your own keys, about a dollar a month
Nothing goes through our servers: judging and translation use your own TypeSafe and OpenRouter accounts, billed by usage, paid to them directly.
- Settings show this month's spend and a monthly estimate from the last 7 days.
- OpenRouter keys can have a credit limit that resets monthly, so you never spend more than you set.
- No keys? The local rules still run.
The developer's own cost for 2.4 days on X, about 5,000 replies viewed (cleaning and translation) — roughly $1.30 a month. The same content isn't paid for twice within 7 days.
Privacy
Nothing is sent to us
No analytics, no tracking
No analytics, crash reporting or ad code. All code is in the package; nothing is downloaded and run later.
Only the two services you choose
The comments and posts you're looking at go to TypeSafe to be judged and OpenRouter to be translated, with your own keys.
Keys and history stay in your browser
Keys are kept in the extension's own database, out of reach of the sites you visit. Uninstalling deletes everything.
Open source
MIT-licensed on GitHub. What is sent, and to whom, is in the code.
FAQ
Why do I need my own API keys?
So what you read never passes through our servers, and there's no subscription to pay or server for us to run. The cost is a few minutes signing up for two services. How to get the keys
Does it work without keys?
Yes. The local rules still clean up referral links, hidden contact details and the same line posted by several accounts. Judging abuse, sexual content and replies like “nice”, and translation, need keys.
Will it hide good comments?
Sometimes. That's why every hide says why, opens with a click, and can be marked as wrong. Thresholds were tuned on each site's own data; if it feels too strict, choose “relaxed” in Settings.
Which sites and browsers?
X, Reddit, YouTube, Xiaohongshu (RedNote) and Hacker News. Chrome 120 or later, plus Edge, Arc, Brave and other Chromium browsers.
Which languages?
The interface is in 19 languages and follows your browser's language; translations go into that language too.
Is it affiliated with X or YouTube?
No. ClearFlow is an independent open-source project, not affiliated with X, Reddit, YouTube (Google), Xiaohongshu or Y Combinator.
How do I report a problem?
Use Feedback in the extension's settings, open an issue on GitHub, or write to developer@ohmyai.app.