
Learn English words while typing: a Rust input method built by one person, 7,600 stars in two months, but the idea was someone else's first
What it is An open-source Pinyin input method. While you type Chinese, each candidate shows an English (or Japanese, Spanish, etc.) word next to it, so you pick up vocabulary as you type.
Why it took off "Development" is understood from a single screenshot. It was first reshared by tech accounts on X and Weibo that repost GitHub projects, then Douyin learning-English and computer-tips accounts turned it into "learn words while typing" videos. Stars rose 823 in a single National Day holiday, and downloads are mostly regular Windows users.
BHow you could follow Don't build another Pinyin input method with translation hints for the Chinese market. Instead, do the mobile keyboard it doesn't handle well (the most-asked request in the comments), or build the same "input method plus a gloss per word" for other audiences like Japanese, Vietnamese, or native English speakers, and promote through tech-repost accounts and short-video accounts.
- Data (updated daily) · Repo created 2026-09-07 · 7,653 stars · 7 days +243
GitHub · stars Data 2026-10-09 → 2026-11-08 (31 days, 31 observations). Points before 10-28 come from deeply / archive backfill, not every day has one
Qingjian is an open-source pinyin input method that adds an English (or Japanese, Spanish) gloss beside each candidate word. It didn't invent this feature: Metasequoia IME did it long before, and Qingjian's own design doc says so. What it did was make this one feature its entire pitch, so clear that a single screenshot explains it. Chinese tech bloggers shared it first, then a Douyin video about "typing while learning words" during the National Day holiday pushed it into programmer circles. Here's how it climbed, who's copying it, and where you can still get in.
A Douyin video on day two of National Day
On Oct 2, 2026, a Douyin account called "Kaiwu" with just 510 followers posted a video opening with: "Learn words while you type?" It went on: "Tried the open-source input method Qingjian on GitHub: while typing Chinese, the candidate words show English definitions and parts of speech, and you can pick Japanese or Spanish too. Writing plans and notes, you pick up a few words along the way." Tags: #EnglishLearning #ComputerTips.
The video got 2,041 likes, 1,860 saves and 474 shares. Saves almost matched likes, meaning viewers were "bookmarking it to install later." That day Qingjian gained 823 stars on GitHub, its best day since the repo was created, ranking #5 across all of Trendshift. It added 582 and 424 over the next two days, and several copycat "install Qingjian with Doubao in one click" tutorials appeared.
The comments show who the audience was: "Is there a phone version?" "Does it work on Apple?" "Does it support Wubi?" "Where do I set it up?" "It's gone, can anyone share the installer?" These aren't programmers. The top comment was a cold splash of water: "Still a bit useless. You can pick up words, but forming sentences and grammar gets hard" (9 likes). Another said "It'd be better as a plugin." As we'll see, someone did exactly that.
What it is
The README's first screen has one line: "Type well, and learn a word along the way." Below it, a candidate box:
``text 1 开发 development 2 编程 programming 3 架构 architecture ``
Only one language's gloss shows at a time: English, Japanese, Spanish, Vietnamese or German, or turn it off. The input method itself is seriously built: whole-sentence input, abbreviated pinyin, error correction, 9 double pinyin, Wubi, and a small local model that reranks whole-sentence candidates after you pause. Everything runs offline, no account needed, cloud suggestions off by default. macOS and Windows are in beta, Linux needs a manually started service, and Android is a preview.
It's mostly one person. The main author on GitHub, yenharvey, has 605 commits; second place has only 26. The repo was created in Sep 7, shipped 14 releases in two months (v0.1.0 through v0.3.2), merged about 200 PRs, and issue numbers have reached #924.
Building a feature like this used to mean paying for dictionary licenses. Qingjian's gloss table was generated offline in bulk by large models: DeepSeek for English, Japanese and German, GPT for Vietnamese. The design doc records that in Sep 5 they ran the whole custom lexicon through, 239k words in total, with one batch of 176k words taking "about 100 minutes and a few dozen yuan."
How the stars grew
Two waves, from completely different sources.
| Timeline | What happened | Stars gained that day |
|---|---|---|
| 09-13 | The author's own Douyin account "Canned Iced Tea" (244 followers) posted the first intro video, 75 likes | +13 |
| 09-14 | @geekbb reposted it on X (50.4k views, 371 bookmarks); the Weibo account "Ant Factory" recommended it the same day | +506 |
| 09-15 ~ 17 | 09-16 @jaywcjlove (小弟调调) reposted it on X, 37.8k views, later added to awesome-mac; there's a thread on linux.do | +273 / +311 / +172 |
| 09-18 ~ 25 | Plateau | 16–61 per day |
| 09-26 ~ 28 | A Douyin video from "GitHub is great," 1,591 likes, 1,518 saves | +159 / +225 / +146 |
| 10-02 ~ 04 | National Day: a Douyin video from "Kaiwu" and a string of install tutorials | +823 / +582 / +424 |
| 10-05 ~ 10 | 10-06 Tencent News and Sohu's "Lao Wang's Productivity Handbook" review | 236–473 per day |
| After 10-11 | slowly declining; 10-30 the author's second Douyin video got 731 likes, with no clear peak | 30–50 per day |
Data: Trendshift daily star gains, GitHub daily star records (11-06 about 7,600 stars, 645 forks; last 7 days +275, the first 7 days +281).
The first wave was programmers. @geekbb, 小弟调调 and Ant Factory post new GitHub projects every day, and tags like Rust, open source and local models hit their sweet spot.
The second wave was ordinary people who want to learn English, and most were on Windows. Look at the downloads: for v0.1.4 alone, the Windows installer got 52,087 downloads, macOS (Apple Silicon) 8,776, and Linux 2,415. All versions combined are about 166k downloads (mixed with update-check requests, so not a user count). Each release's changelog also links a Quark Drive download, so Douyin users can install without visiting GitHub, and that volume is invisible from outside. Download sites like ddooo and ghxi have already reposted it.
It never got on Hacker News, never launched on Product Hunt, and is basically unknown in English-speaking tech circles. On X there are a few near-identical recommendations ("I type hundreds of pinyin a day, didn't know I could learn English too"), with only a few hundred views each, looking like a content farm repost. Whether they were paid is unverified.
Others built the feature first. Why did this one take off?
Metasequoia IME (MetasequoiaIME) already had translations beside candidates on Windows in October 2023. Qingjian's design doc docs/design/landscape.md is candid: "The 'see the translation while typing' interaction has already been validated by Metasequoia, so no prototype validation is needed," and says Qingjian's edge over Metasequoia "comes down to three things": macOS and Linux support, a platform-agnostic Rust core, and showing only one gloss (Metasequoia shows up to two, and can use the cloud).
But Metasequoia treated translations as one feature among many, while Qingjian made it its whole identity: the name, the slogan, the first README screen and the demo video all talk about this one thing. That's what content accounts need: one "development: 开发" screenshot that explains itself in 15 seconds, no need to understand input methods. Layer on the long-running complaints about Chinese input methods (too many ads, always online, poor privacy), and "clean, no account, fully local" becomes the second selling point almost for free.
Two bits of evidence. Metasequoia has more features, and Qingjian's docs record it at about 950 stars in September. By October, it moved "type Chinese, see English beside the candidates" to the first screen of its README and added a vocabulary-book feature, effectively learning from Qingjian. Another is Glint from late October, which copied only this one point and got 119 replies on V2EX.
Who it's competing with
| Project | Launched | Size (11-06) | How it's done |
|---|---|---|---|
| Metasequoia msime-windows / msime | 2023-10 / 2026-09 | 3,357 stars / 1,167 stars | Added translations beside candidates first, promoted it to the main selling point from October |
| Rime + Rime Ice (雾凇拼音) | Years; rime-ice 2021 | 20K stars | You can get the same effect with a comment plus a Lua script; some users already moved Qingjian's lexicon into rime-ice |
| Glint(jsogn/Glint) | 2026-10-20 | 98 stars, V2EX thread with 119 replies | Doesn't replace the input method; reads the system IME's candidates and overlays a translation popup on top. macOS only |
| GlintWin | 2026-10-31 | 3 stars | Windows version of Glint, uses OCR to read Sogou's candidates |
| Glimmer (微明) | 2026-09-15 | 53 stars | A fork of Qingjian; drives traffic through links under 小弟调调's posts, and set up its own site |
| hallelujahIM | 2014 | 2,601 stars | The reverse: English IMEs showing Chinese glosses |
Big-name IMEs (Sogou, Baidu, iFlytek) have translate-on-commit modes, but I haven't seen the "a gloss always visible beside every candidate" learning positioning (not verified one by one).
So the pattern is visible: in Chinese tech circles, the "learn words while typing" idea was copied three or four times within a month, and the copycats got only dozens to a hundred stars. The real barrier is the input method itself. Users only switch if it's "no worse than what they already use."
If it were you
Verdict: Building another pinyin IME with glosses in the Chinese scene is too late. But the approach of "slipping learning into an action you do hundreds of times a day, and narrowing the pitch to one screenshot" hasn't been done yet for other audiences and platforms.
Who it suits: indie developers who can use AI to build a complete, usable system-level tool in a month or two; people who know the input habits of a particular language community (Japanese, Korean, Spanish, Vietnamese users); or people already familiar with Rime or mobile keyboard extension development.
Who it doesn't suit: anyone who only wants a translation overlay and expects thousands of stars (Glint did that, 98 stars); anyone unwilling to polish the input method itself to daily-use quality.
How to do it: three openings, ranked by confidence:
- Mobile. The most common comments were "Is there a phone version?" and "Does it work on Apple?" Qingjian's Android is still a preview and there's no iOS. A Chinese keyboard with glosses for iOS / Android is the part it hasn't done well.
- Other native-language audiences. Japanese people learning English, Vietnamese people learning Chinese, English speakers learning Chinese: an input method plus one gloss hasn't been seen in those communities. Qingjian's gloss table cost a few dozen yuan of large-model runs, and the data cost is just as low for other languages.
- A plugin. Build an out-of-the-box "learn English" scheme for Rime / rime-ice, or a solid overlay for Sogou users on Windows (GlintWin has only 3 stars and isn't good yet).
How to get users: follow the two steps Qingjian took. First, make the README's first screen a screenshot you understand at a glance plus one slogan, and send it to the tech accounts that repost new GitHub projects every day (for Chinese, the handful on X and Weibo; for other languages, find the equivalent repost accounts). Then get small accounts posting "learn a language" and "computer tips" on Douyin / TikTok / YouTube Shorts to film 15-second demos. Offer a Drive or store download for people who don't use GitHub. Post around long holidays: Qingjian's biggest wave came during National Day.
How early you need to be: the Chinese route is already closed; the mobile keyboard route and the other-language routes probably have three to six months left, and they narrow once Metasequoia or Qingjian ships a mobile version itself.
What would mean you misread it: your version ships, tech accounts repost it, and after a month it still has only a few hundred stars; or Qingjian's Android graduates and an iOS version appears; or a big-name IME launches always-on glosses.
Research notes: competitor table, acquisition evidence, full research report
- Competitors
Name Where Launched Scale How it differs from this one Metasequoia IME (MetasequoiaIME) Open source Windows version 2023-10; multi-platform Rust version 2026-09-02 msime-windows 3357 stars, msime 1167 stars Showed translations beside candidates earlier, but only on Windows at the time; up to two glosses, can use the cloud; a feature among many. Qingjian made it the sole positioning, supports macOS, shows one gloss, fully offline Rime + Rime Ice (雾凇拼音) Open source Years; rime-ice 2021 rime-ice 20k stars A comment plus Lua can produce similar annotations, but you have to configure it yourself; no out-of-the-box learning positioning hallelujahIM Open source 2014 2601 stars English IMEs showing Chinese glosses, the opposite direction Sogou / Baidu / iFlytek / WeChat IME Big-company features Years Hundreds of millions of users Have translate-on-commit modes, but no always-visible gloss positioning for learning (not verified one by one); Qingjian contrasts itself on no ads, offline, no account Glint / GlintWin Open source 2026-10-20 / 2026-10-31 98 stars / 3 stars Doesn't replace the IME; overlays a translation popup on the system or Sogou candidates. A copycat approach - Where users come from
- Verified Chinese tech accounts on X: @geekbb posted 09-14 tweets, 50.4k views, 371 bookmarks; @jaywcjlove posted 09-16 tweets, 37.8k views, and got added to awesome-mac; stars that day: +506 / +311
- Verified Weibo: Ant Factory 09-14 post recommendation (reposted on sina.cn)
- Verified Douyin content accounts: "GitHub很棒" 09-26 video 1591 likes, matching star growth of 09-26~28; "开物" 10-02 video 2041 likes and 1860 saves, +823 that day; 10-03 and 10-04 multiple "Doubao one-click install" tutorials followed
- Verified Founder's own content: founder's Douyin "罐装冰红茶" (244 followers) posted a first video on 09-13 with 75 likes; a second on 10-30 with 731 likes
- Verified Communities and download sites: linux.do 09-17 threads; Tencent News / Sohu reviews 10-06; CSDN tutorials; ddooo, ghxi and other download sites reposting; YouTube/Instagram re-uploads
- Verified Quark Drive distribution: each release changelog includes a Quark Drive link, so Douyin users who don't use GitHub can still install it
- Inferred Templated promo accounts on X: @bkdgiffug, @Huahuazo and others use similar copy patterns with only a few hundred views, likely a batch of content-farm reposts
Why Qingjian (qingjian-team/qingjian) broke out
Research date 2026-11-06. Channel: GitHub.
One line
Qingjian is an open-source Rust pinyin input method (macOS / Windows, with Linux and Android in preview) that shows one extra line of "the foreign word you're learning" next to each candidate. It wasn't the first input method to show translations beside candidates (Metasequoia IME came earlier, as Qingjian's own design doc acknowledges), but it made this the product's single selling point, backed by an easy-to-grasp demo video ("开发 development"). Chinese tech accounts on X (@geekbb, @jaywcjlove) and Weibo's Ant Factory reshared it first, then Douyin "type while memorizing words" videos during the National Day holiday carried it from the programmer crowd to general users: Windows installer downloads are 6 times those of macOS.
1. Basic data
| Metric | Value | Source |
|---|---|---|
| Repo created | 2026-09-07 (organization created 2026-09-06) | GitHub API |
| Star / Fork | 7,628 / 652 (live as of 11-06); library records 11-06 / 7,600 / 645 | GitHub API; github database repo_day |
| Stars gained in the last 7 days | +275, +281 in the first 7 days, averaging 30–50 per day, cooling off | Data card; GitHub repo |
| Issue number | Reached #924, 206 open | GitHub API |
| Commits | Lead author yenharvey with 605 commits, runner-up with 26; 200 PRs merged in 56 days | contributors API;trendshift |
| Releases | 2 months, 14 app releases (v0.1.0 to v0.3.2) + 9 data packages | releases API |
| Release downloads | Total app release downloads about 166k (including update-check files like SHA256SUMS and releases.json, so not a user count). v0.1.4 alone: Windows installer 52,087, macOS arm64 8,776, Linux 2,415 | releases API 2026-11-06 |
| Other distribution | Each release includes a Quark Drive link (so domestic users don't need GitHub); ddooo, ghxi, 32r, china.com and other download sites have reposted it | README; search results |
Star growth curve (Trendshift daily data, 2026-09-12 to 11-06)
| Date | Stars gained that day | Events in the same period |
|---|---|---|
| 09-13 | +13 | Founder's own Douyin post "I wrote an input method in Rust" (75 likes) |
| 09-14 | +506 (Trendshift site-wide #20, Rust #3) | X @geekbb post (50.4k views, 348 likes, 371 bookmarks); Weibo "Ant Factory" recommendation the same day; Douyin "请叫我刘小猫" (88 likes) |
| 09-15~17 | +273 / +311 / +172 | 09-16 X @jaywcjlove (小弟调调) post, 37.8k views; 09-17 linux.do thread (540 views) |
| 09-18~25 | 16–61 per day | Plateau |
| 09-26~28 | +159 / +225 / +146 | Douyin "GitHub很棒" (8,453 followers) video with 1,591 likes and 1,518 saves |
| 10-02~04 | +823 / +582 / +424 (10-02 Trendshift site-wide #5) | National Day holiday. Douyin "开物" (510 followers) video with 2,041 likes, 1,860 saves, 474 shares; 10-03 and 10-04 more "install Qingjian with Doubao in one click" and "input method install tutorial" copycat videos |
| 10-05~10 | 236–473 | 10-06 Tencent News / Sohu "老王效率手册" review; 10-05 X @hank_aibtc and others sharing their experience |
| After 10-11 | Gradually cooled to 30–50/day | 10-30 founder's second Douyin post "my open-source input method evolved" with 731 likes, no clear star spike |
Takeaway: two peaks. The first, 09-14~17, came from Chinese tech accounts on X and Weibo, aimed at programmers; the second, 09-26 and 10-02~08, came from Douyin "typing while memorizing words" content accounts, aimed at ordinary users who want to learn English (mostly Windows users). The second peak was bigger.
II. Competitors
| Product | Where | Launched | Scale | How it differs from Qingjian |
|---|---|---|---|---|
| Metasequoia IME MetasequoiaIME (msime-windows / new msime) | Open source | Windows version 2023-10; Rust multi-platform version 2026-09-02 | msime-windows 3,357 stars, msime 1,167 stars (GitHub repos 11-06); Qingjian docs record September ~950 stars and "5000+ new users recently" | Built "translations beside candidates" earlier than Qingjian (Qingjian's docs/design/landscape.md says: "this interaction form has already been validated by Metasequoia"). But at the time Metasequoia was Windows-only, showed at most two translations, could use cloud translation, and this was one feature among many, not the main selling point. From October, Metasequoia's README moved "type Chinese, see English beside the candidates" to the first screen and added a vocabulary book, effectively converging on Qingjian's approach. |
| Rime (Squirrel / Weasel) + Rime Ice (雾凇拼音) | Open source | Over a decade; rime-ice 2021 | rime-ice 20k stars | You can add translation annotations with the comment field and a Lua script, but you have to set it up yourself. Users on X have already "added Qingjian's dictionary to Rime Ice" and reproduced the same effect |
| hallelujahIM English input method | Open-source macOS | 2014 | 2,601 stars | Opposite direction: an English input method showing Chinese glosses |
| Sogou / Baidu / iFlytek / WeChat input methods | big tech | Years | Hundreds of millions of users | Has a Chinese-English translate-on-commit mode, but no "a translation line permanently attached to every candidate" learning positioning (not verified case by case); Qingjian users' complaints were precisely about "ads, networking, privacy" |
| Toucan (acquired by Babbel) and other immersive vocabulary browser extensions | Chrome | Around 2019 | Not verified | Also "learn words incidentally in daily activity," but it happens while reading web pages, not while typing |
| Copycat: Glint (jsogn/Glint) | Open-source macOS | 2026-10-20 | 98 stars; V2EX thread 119 replies (10-29) | Doesn't replace the input method; reads the system input method's candidates and overlays a translation popup |
| Copycat: GlintWin | Open-source Windows | 2026-10-31 | 3 stars | Inspired by Glint, uses OCR to read Sogou candidates and overlays English |
| Copycat: Glimmer (微明, OpenFabrica/glimmer) | Open-source, a fork of Qingjian | 2026-09-15 | 53 stars | Posted its own fork as a referral under jaywcjlove's post, and set up a separate site glimmerinput.app |
Judgment: "translations beside candidates" was not invented by Qingjian; Metasequoia did it first, and Rime can be configured to do it. Qingjian's difference is that it made this the sole positioning (the name, slogan, README first screen, and demo video all talk only about this), covers macOS, and shows just one line, fully offline, with no popups, deliberately restrained. It was this single-point, easy-to-explain positioning that let content accounts explain it in 15 seconds, and that is why it broke out.
III. Growth levers (where users come from)
- 中文科技 X 大号与微博(已确认):09-14 @geekbb(5.04 万浏览、371 收藏)、09-16 @jaywcjlove(3.78 万浏览,后来还把青简收进了 awesome-mac)、09-14 微博「蚁工厂」。这些号日常就在搬 GitHub 新库,青简的 README 和演示视频天然适合被搬。
- Douyin "type while memorizing words" content accounts (confirmed, and the largest wave): "GitHub很棒" (09-26,1,591 likes), "开物" (10-02,2,041 likes, 1,860 saves), plus copycat tutorials from "风口的居", "肖老二", "达人AI小华" and others. Saves are close to likes, meaning users were "bookmarking to install later," a typical trait of tool content. The timing fell right on the National Day holiday.
- Founder posting content directly (confirmed but small in volume): the founder's Douyin account "罐装冰红茶" (244 followers, usually posts about AI / cosplay / vibe coding) posted a first intro video (75 likes) on 09-13, and a second one on 10-30 (731 likes). The video copy reads like product-manager writing and includes three entry points: website, GitHub, and Quark Drive.
- Community threads and download sites (confirmed): linux.do (09-17), mentioned in several V2EX threads, reviews and tutorials on CSDN / Tencent News / Sohu, and reposts on download sites like ddooo and ghxi. There are also auto-narrated YouTube channels in the "GitHub Explained" style and English Instagram reels (captioned "4k+ GitHub Stars"), which fall into the long tail of re-uploads.
- Quark Drive distribution (confirmed): each release changelog includes a cloud drive link, so Douyin users can install without visiting GitHub. This also explains the real install numbers that followed the star growth.
- Suspected templated promo on X (unconfirmed): @bkdgiffug "hitting pinyin hundreds of times a day, didn't expect to pick up English on the side" and @Huahuazo "typing pinyin dozens of times a day, somehow learning English too". The copy patterns are very similar and views are only in the hundreds. Could be a content farm reposting in bulk, or paid promo; no evidence of paid spend found.
- Not found: not on HN or Product Hunt (our hackernews and producthunt databases return nothing), and not featured in Ruan Yifeng's weekly issues 410–418. No evidence of ad spend either.
4. Why it made the list
- Explainable in one line, readable in one screenshot: a candidate-window shot like "开发 development" needs no explanation. Content accounts love this kind of asset.
- Hits two stacked mass-market pain points: wanting to learn English without spare time ("hundreds of keystrokes a day"), plus distrust of domestic IME apps (ads, online connectivity, privacy). One tool covers both.
- Borrows developer-circle trust, then gets amplified by mainstream accounts: the first wave came from tags like Rust, open source and local models, which pulled developers on X and Weibo. The second wave on Douyin cared more about "vocab learning", and Windows downloads far outnumbered macOS (v0.1.4:52,087 vs 8,776), showing users had left the developer circle.
- Fast iteration, clearly a serious input method engine: 2 months, 14 releases, with whole-sentence input, 9 dual-pinyin schemes, Wubi, error correction, and a local ranking model. Issues are up to #924. Reviews repeatedly call it "unobtrusive" and "restrained," which suggests the experience keeps people around.
- Extremely low cost for a solo dev plus AI: the main author made 605 commits solo; the translation word list was batch-generated offline with DeepSeek / GPT, 239k words in about 100 minutes for "a few dozen yuan" (docs/design/landscape.md). This solves what used to be the most expensive step: where the translation data comes from.
5. Differentiation, and whether it's why it took off
The differentiation isn't the feature itself (Metasequoia shipped it first). It's positioning and packaging:
- Turning "learn a language on the side" from one feature among many into the product's entire identity (name, slogan "Type well, and pick up a word along the way", first screen, demo video);
- Showing only one translation at a time, no quizzes, no interruptions, fully offline, with multiple language options (English / Japanese / Spanish / Vietnamese / German);
- Available on both macOS and Windows at once, when Metasequoia was Windows-only at the time;
- Privacy promises up front: no account, cloud features off by default, and you must enter your own AI provider to turn them on.
This differentiation is why it took off: content spread rode the "type and memorize words" hook, not Rust and not the whole-sentence model. Two side signals: Metasequoia had more features but didn't take off until October, when it moved the same selling point to the first screen; and Glint and GlintWin, which appeared in late October, copied only this one point and still got 2 replies on V119EX.
6. How many people have seen it
- 7.6k stars, two Chinese X accounts reposted it (about 90k views combined), a top Douyin video with 2 thousand likes, plus coverage from Tencent News and Sohu reviews. It's already well known in Chinese dev circles and the "productivity tools / learn English" Douyin crowd, but almost nobody in English-speaking circles has heard of it (no HN or PH posts).
- Free, so there's no public revenue to report.
- Copies already exist: Glint (98 stars), GlintWin, a Glimmer fork, and Metasequoia is also reinforcing the same selling point. The idea is easy to copy; the real moat is the quality of the IME core.
VII. What others can borrow
- "Put learning inside behavior that already happens frequently": input methods, browsers, clipboards and notification bars are all candidates. Narrow the pitch to one sentence plus one screenshot.
- Use an LLM to batch-generate content data offline (translations, parts of speech, pronunciations): a few dozen yuan buys a dictionary that used to require a license.
- Get stars and trust first from Chinese tech X / Weibo repost accounts, then let Douyin tool accounts drive mass reach; add Quark Drive downloads to catch users who don't use GitHub.
- When a competitor hasn't made a feature its selling point, packaging that single point is the opening (that's how Qingjian went after Metasequoia).
- The flip side: single-point packaging gets copied quickly (Glint appeared within a month), so you need core quality and iteration speed to defend it.
VIII. Unknowns
- Real active users: GitHub download counts include update-check files, and Quark Drive volume isn't visible; how many of the 7.6k stars came from Douyin users is unknown.
- No evidence whether the templated copy on X is organic reposting or paid promotion.
- Specific engagement numbers for the Ant Factory Weibo post weren't obtained; Xiaohongshu and Bilibili volumes weren't checked (the TikHub Xiaohongshu API was unavailable).
- The exact launch date of Metasequoia's "candidate side translation" wasn't found; only Qingjian's doc note "implemented as of 2026-09" is available.
- Whether major-vendor IMEs have a similar always-on translation feature wasn't verified company by company.
Sources
- GitHub repo and API: github.com/qingjian-team/qingjian (README, releases, contributors, docs/design/landscape.md)
- Trendshift daily star growth and ranking: trendshift.io/repositories/233641
- Our github repo_day / repo_profile tables (2026-10-09 to 11-06)
- X:x.com/geekbb/status/2099343127672750267 ;x.com/jaywcjlove/status/2100072596096286970 ;x.com/jaywcjlove/status/2100196358057853158 ;x.com/hank_aibtc/status/2106947208432267567 ;x.com/Huahuazo/status/2111571453489647681
- Weibo Ant Factory (via sina.cn): www.sina.cn/news/detail/5343023516877859.html
- Douyin (data via TikHub API): videos 7692101967908195593 (Kaiwu), 7689773783904832802 (GitHub is great), 7702439848900169011 / founder account "Canned Iced Tea", 7685390847688363310 (Call me Liu Xiaomao)
- linux.do:linux.do/t/topic/2914721
- Tencent News review: news.qq.com/rain/a/20261006A06DIO00
- V2EX (API): www.v2ex.com/t/1253095 (Glint), www.v2ex.com/t/1253914 (GlintWin)
- Competitor repos: metasequoiaime/msime, metasequoiaime/msime-windows, jsogn/Glint, fengwuyun/GlintWin, OpenFabrica/glimmer, iDvel/rime-ice, dongyuwei/hallelujahIM