Stringers
BigPizzaV3/ClaudeMom · BigPizzaV3

Hypnotizing Claude Desktop: a library split out from a 33k-star old project. 6 days, 600 stars

What it is A desktop tool that fakes the official server on your own machine, telling Claude Desktop that you're logged in. People without a Claude account can use all the logged-in features with their own API key.

Why it's blowing up The official April rollout cut back the no-account mode, and Chinese users are afraid of getting banned, so this gap has been there for a while. The author split a newly shipped feature from Codex++ (33k stars) into its own library. The existing user base and WeChat Official Accounts drove the first wave in the Chinese-speaking community, then a big Russian Telegram channel amplified it again with a 'hypnotizing the official client' angle, 11-16.

CHow you could follow Don't copy this feature itself; copy the playbook. Split the most viral feature of a big tool into its own library, use a one-line hook as the description, and put a three-column comparison table in the README: "official full version / official stripped version / you." Look for the next opportunity in the gap between the official degraded mode and the full version.

Grade C: the traction came from the author's 33k-star old project, existing code, and community, none of which a follower can get. The wave has already peaked, the total is only about six hundred stars, few open niches remain, and it carries the risk that an official update blocks it at any time or that it receives a DMCA takedown. Worth learning the playbook, not worth copying the project. Whispers in small circles
Data (updated daily) · repo created 2026-11-12 · 646 stars
GitHub · stars
025050075011-1511-1611-1711-1811-1911-20646
Data 2026-11-15 → 2026-11-20 (6 days, 6 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

ClaudeMom makes the official Claude desktop app think you're logged in. As a result, features only paid accounts get (computer control, voice, the full Cowork, and the Chinese interface) are unlocked, and model requests actually go through your own API key. It was created 11-12, and as of 11-18 it has 591 stars and 122 forks. Below: what it actually does, why it caught fire this particular week, where the stars came from, and what you can and can't borrow from it if you want a project that gets stars too.

A post on a Russian channel

Nov 16, noon Moscow time: the Russian-language Telegram channel "Вайб-кодинг" (81k subscribers) posted:

"I came across a completely insane project on GitHub called ClaudeMom. At first I thought the author had just built a knockoff Claude client. Reading the source showed how wrong I was: he isn't swapping in a fake Claude. He straight up hypnotized the original."

A later paragraph explains the mechanism: when the official client checks with the server for login, a local fake gateway answers first with "logged in, all permissions on," and the client believes it. The post got 21.7k reads. That same day, the news aggregator gotburnout translated it into 6 languages under the title "Hypnotizing the official client." Kazakhstan's AI Qazaqstan translated it into 4. The next day (11-17), ClaudeMom gained 326 stars in a single day, its highest single-day total so far.

But the Russian-speaking community was only the second wave. The project was written by a Chinese developer, its README is only in Chinese, and the earliest users were in China.

What it actually does

Claude Desktop has two modes: log in to an official account for the full feature set, or turn on "developer mode" and enter a third-party API URL and key (called 3p mode below). The second works without an account, but many entry points are gone. ClaudeMom is a third option: it installs a certificate and runs a local service on your machine, answers on behalf of the official server that "this person is logged in and these features are on," and forwards the actual conversation requests to the upstream you configure (a relay station, or DeepSeek or any endpoint compatible with the Anthropic API). It doesn't modify the official app, which keeps updating as usual.

The most viral part of the README is this table (verbatim):

Log in to official accountAPI key (3p mode)ClaudeMom
Requires Claude accountYesNoNo
Model sourceOfficialYour keyYour key
Chat / Code / Cowork✅Partial✅
Computer control / voice input✅❌✅
Chinese interface✅ but I don't dare turn it on❌✅

"✅ but I don't dare turn it on" means nothing to outsiders, but Chinese users get it instantly: people with accounts worry that switching the interface to Chinese could be a signal that gets them banned. A line only the target users understand makes them feel "this was made for me," more than a full paragraph of feature description would.

It's a finished desktop app written in Tauri + Rust. The macOS build is signed and notarized, so it opens on download. It has auto-updates, can import configuration from cc-switch (a provider-switching tool with 155k stars) in one click, and its voice input uses a local offline model, so audio never leaves the machine.

Why this week: the official side left an opening

This demand wasn't created by ClaudeMom. The official side cultivated it half a year ago.

  • April 2026: Claude Desktop opened 3p mode, and Chinese-language feeds filled up. AI bloggers on X and the Tencent Cloud developer community both published tutorials. "Running the official client on a relay key" became a daily habit for many people.
  • September: reports said 3p mode added a model allowlist, blocking non-Claude models.
  • Meanwhile, 3p has always lacked computer control, voice, and part of Cowork, and it has no Chinese interface. Getting the full feature set requires an account, and Chinese users are most afraid of getting banned.

So a group of people got stuck in the middle: they have a key, but no account or don't dare use one, and they want all the features of the logged-in version. ClaudeMom's description opens by speaking directly to them: "No Claude account needed! Turn your API key into logged-in mode." The WeChat reposts that followed mostly copied this line for their headlines.

How the stars climbed

DatestarAdded that dayforkWhat happened these past few days
11-12———Created the repo and released v0.1.0 and v0.1.1 the same day; that evening posted in the "open-source promotion" section of linux.do
11-13———First WeChat repost ("南京Sol")
11-1522—0First fork in the afternoon (Beijing time); forks kept climbing after that
11-16180+15828Full-article repost on linux.sb; Russian channel post; the author shipped 4 versions in one day
11-17506+326102Aggregator sites repost in multiple languages; WeChat accounts keep reposting (8 articles on the first page of Sogou WeChat search)
11-18591+85122Cooling off

Source: GitHub daily star/fork records (from 11-15); the GitHub page shows 11-18 with 596 stars and 124 forks.

A few details:

  • The earliest users were Chinese. On 11-15 there were already 37 forks, and 11-16 forking was densest in the Beijing morning, all before the Russian post. Where they came from isn't directly visible; the most likely source is the author's older project's QQ and Telegram groups (inference).
  • The linux.do promo post itself drew little traffic: about 424 views and 15 replies. Opening it now shows 404, cause unknown. Its role was more like "following the rules": the README also includes a LINUX DO friend link, per community custom.
  • Fork rate is about 20%. New repos of similar size typically sit around 11% (estimate). This is common for Chinese tool repos: people fork first, afraid it will get taken down.
  • The people who actually installed it number roughly a thousand: as of 11-18,9 versions, the macOS package has been downloaded 308 times and the Windows package 692 times (GitHub Releases).
  • The English-speaking world barely moved: no English tech media coverage, no big X accounts, and no HN or Reddit discussion threads found.

The real trump card: the author's 33k stars

ClaudeMom didn't start from zero. The author BigPizzaV3's other project, Codex++ (an enhancement tool for Codex), has 33k stars, a 7,400-member Telegram group, and a QQ group on its 4th iteration. Zhihu has an article titled "Hacking Codex: the enhanced Codex++ is blowing up on GitHub!"

ClaudeMom's core code is the "Claude full-feature mode" that just shipped in Codex++ v2.6.0 11-10. Two days later the author split it out into its own repo, and the first commit reads "split out from CodexPlusPlus"; the same day, the Codex++ release notes already mentioned ClaudeMom.

So from the first day it was a finished, installable, signed product, backed by a group of old users who trust the author. The issue tracker shows the same relationship: over 6 days the author replied to issues one by one, and an external developer named wuwuzhazha opened 8 issues and fixes on Windows and Cowork (4 of them with code attached). One user wrote in an issue that they had "reverse-engineered the Design and Slide features with DeepSeek" and were "just waiting for the expert to build this library."

This is the thing a follower can't get.

Who else is doing the same thing

ProjectScaleHow ClaudeMom differs
Official 3p modeEveryone who has the desktop app installedOfficially supported, no certificate needed, but missing a chunk of features
cc-switch155k starsOnly helps you write 3p config, doesn't fake login state; ClaudeMom can import from it, so the two complement each other
claude-desktop-zh-cn8,655 starOnly a Chinese localization patch
claude-desktop-multiple-windows-different-api985 starMultiple windows, one on subscription and one on 3p; requires an account
claw-in-chrome527 starThe same idea applied to Claude's Chrome extension, already available since April
openwork, eigent, AionUi, etc.15k to 34k starsRewrite a client from scratch instead of using the official one

star count from GitHub, 11-18 before and after.

"Use without logging in" was already done by the official client and cc-switch, and "Chinese localization" already has patches with 8,600 stars. What ClaudeMom adds is faking the login state, and that is the core line of all its promotional copy, and also the part that Russian-speaking users find "crazy." The only copycats so far are reskinned forks: one renamed itself and changed its trilingual (Chinese, English, Russian) intro, and another publishes its own releases with a higher version number than upstream. No independent rewrite has appeared yet.

The shelf life of this craft

Spoofing a login is a race against the official client. The issue tracker already shows the cracks: 11-17 someone reported that after opening it, the settings "change" again after a while, and others said Chat messages all fail to render on the new Windows build. The author shipped 9 releases in 6 days, much of it chasing changes in the official client. Whether Anthropic will close this path with a client update, or send a DMCA takedown, nobody knows yet. The star curve also says this wave has peaked: 11-17 +326,11-18 only +85.

If it were you

C: Worth learning from, but following it isn't worth the effort. If you replicate ClaudeMom, you won't have the author's 33k stars of established users or the existing code. You'd be a week late, and this wave has already peaked, with only about six hundred stars total. Doing the same thing with a different AI client leaves little room (Codex was taken by the author himself, and a Chrome extension called April has already been built). You'd also carry the constant risk of being shut down by the vendor or hit with a DMCA, and you'd be chasing client updates every day.

Who it suits: developers who already have a tool with a few thousand stars and their own user base, and who know the internals of some AI desktop client. You'd be learning the author's playbook, not the feature.

Who it doesn't suit: anyone who wants to grow stars from scratch with a new library; anyone who can't stand patching overnight after every official update; anyone who cares about legal risk and doesn't want their GitHub account tied to a DMCA.

How to do it: copy three things, not the feature itself.

  1. Split the most viral feature of your big tool into its own library, give it a memorable name, and write the description as a single hook line. The feature has already been proven in an established project, and once extracted it has a chance to be shared on its own.
  2. Look for the gap between the official "stripped-down" and "full" versions. If the official offers a login-free, API-based or free degraded mode, the community will want "the low-barrier degraded mode plus the full feature set."
  3. Put a three-column comparison table in the README (official full version / official degraded version / yours), and add one line that only the target users will understand. Ship release one as a finished product: signed, with auto-updates, and able to import config in one click from the tools people already use.

If you insist on this direction, the thing to build is what it does poorly. Issues already ask for an Ubuntu version, and for Design and Slide support, but these are all tied to the author's code and pace. Building them is more likely to produce PRs to the original than a new project that can gather stars on its own.

How to get users: follow its route in the Chinese-speaking community. Build up the existing user base first, post promotional threads on linux.do while following its rules, and write a headline that WeChat Official Accounts will happily copy. To cross languages, projects with a technical twist ("tricked the official client") are most easily picked up as curiosities by Russian and English Telegram news channels, so you can actively pitch to those channels.

Deadline: there isn't really one. The window for these "unlock" projects is set by the official vendor and often moves on a weekly cycle. To reuse this playbook, launch within the first week after the official releases a stripped-down login-free or API mode.

Signs you misjudged it: (1) two weeks later ClaudeMom is still gaining hundreds of stars a day, or an independently rewritten similar project also reaches several hundred stars, which would mean demand is bigger than it looks and worth reassessing; (2) conversely, if an Anthropic update breaks it or the repo receives a DMCA, the value of even studying this path drops; (3) if someone carries the same idea over to another big vendor's AI client and gets several hundred stars within a week, the "gap between stripped-down and full versions" playbook is replicable and worth grabbing the next one.

GitHub repo ↗

Research notes: competitor table, acquisition evidence, full research report
Competitors
NameWhereLaunchedScaleHow it differs from this one
Claude Desktop official third-party inference (3p) modeBig-company features2026-04Everyone with the desktop app can use it; dozens of Chinese tutorialsOfficially supported, no certificate install, but no computer control, voice, or some Cowork features, and no Chinese interface; reports since September say non-Claude models are restricted
farion1231/cc-switch (Claude Desktop panel)Open source2025-08 (Claude Desktop panel about 2026-05)155,198 starOnly writes 3p configs and does model mapping, doesn't fake the login state, and is as restricted as 3p. ClaudeMom can one-click import providers from it, so the two complement each other
lonr-6/cc-desktop-switchOpen source2026-04-24804 stars, stalledOnly configures 3p providers
javaht/claude-desktop-zh-cnOpen source2026-04-248,655 starOnly does a Chinese localization patch, doesn't solve account or feature issues
TonF1917/claude-desktop-multiple-windows-different-apiOpen source2026-10-14985 starRun multiple windows, one on the official subscription and one on 3p, provided you have an account
BigPizzaV3/CodexPlusPlus's "Claude full-feature mode"Open source2026-11-10A feature inside a big tool with 33,192 starsSame author, same core; ClaudeMom is a lightweight standalone version that only handles Claude
S-Trespassing/claw-in-chromeOpen source2026-04-05527 starSame idea applied to the Claude in Chrome extension, removing login and plan restrictions; ClaudeMom does it on the desktop app
Open-source Cowork alternatives such as openwork / eigent / AionUiOpen source2025-202615k to 34k starsRewrite the client from scratch; ClaudeMom uses the original client, so its interface and features match the official one
Where users come from
  • Inferred Codex++ author's longtime users and community: Codex++ has 33,192 stars, the Telegram group has 7,422 members, and the QQ groups have grown to 4 groups; the Codex++ v2.6.7 release notes mention ClaudeMom; the Chinese fork wave starting 11-15 09:38 UTC most likely came from here (not visible in group chats)
  • Verified linux.do open-source promo thread: topic 3166605,11-12 posted late, about 15 replies and about 424 views, now 404; the README includes a LINUX DO friend link, following community convention
  • 查实 微信公众号搬运:搜狗微信第一页 8 篇,11-13 到 11-17:南京Sol、慢慢长大的糖糖、光谋纪元、小锋学长、每周AI拾光、泡学堂 GitHub 每日热点速递、陈嗣俊等
  • Verified Chinese AI community reposts: linux.sb/topic/32684 (11-16), 仙人饼社区 (11-12)
  • Verified Russian Telegram and multilingual aggregators: t.me/vibecoding_tg/4355 (81,446 subscribers, 11-16 09:12 UTC, 21.7K reads); gotburnout across 6 languages, AI Qazaqstan across 4 languages, Vibe Coding Daily reposts; this happened on the two days with the highest single-day star gains
  • Verified GitHub project aggregators: HotRepo, Kapyn, Context7, SkillsLLM, and Nicheloom all list it
  • Inferred Big X accounts: neither x-radar nor Brave turned up posts about ClaudeMom

Why BigPizzaV3/ClaudeMom could take off

Data as of 2026-11-18. Star and fork curves come from our GitHub database (repo_day); other sources are linked inline.

Bottom line

ClaudeMom was originally a feature inside Codex++ (same author, 33k stars), which the author split out into its own standalone repo. It targets a clear audience: people in China who use relay APIs, who fear account bans, or who have no Claude account at all but want the full features of the official Claude desktop app. The method is to install a certificate on the local machine and run a local service that tells the client it's already logged in. The computer control, voice, full Cowork, and Chinese interface that the official third-party mode (3p) doesn't provide all come back.

Spread happened in three steps. 11-12 The author posted an open-source promo thread on linux.do; the circle was small. 11-13 From 11-17, a batch of WeChat Official Account reposts followed. 11-16 A large Russian-language Telegram channel posted about it, framing it as a bizarre project that "hypnotized the official client," with 21.7K reads, and multilingual aggregators then picked it up. Stars went from 11-15 at 22 to 11-18 at 591, with the peak at 11-17 gaining +326 in a single day. Since 11-18 it has only gained 85, and it is already cooling off.

What it is

  • A desktop app written in Tauri + Rust. The core is crates/claude-local: a local compatibility service (Node script), a local CA certificate, profile switching, and local dictation.
  • How it works (README wording): "Runs a local compatible service that takes over the Claude desktop app's account and feature requests: account info and feature flags are returned as 'logged in,' and conversation requests are forwarded to the upstream you configured."
  • The selling point is the three-column comparison table in the README: official login / official API Key (3p) mode / ClaudeMom. The ClaudeMom column needs no Claude account, uses your own Key for the model, and ticks every box: Chat, Code, Cowork, computer control, voice, and the Chinese interface. In the official-login column, the Chinese interface entry reads "✅ but I don't dare turn it on," meaning users in China worry that switching to Chinese is a signal for account bans. That line hits the target audience hard.
  • Extras: you can one-click import providers from cc-switch and use Magpie's local gateway as an upstream; offline local dictation uses SenseVoice; the Code tab and the command line share ~/.claude; there are auto-updates; the macOS build is signed and notarized.
  • Source: README

Background (timeline)

TimelineEventSources
2026-04Claude desktop opens "Developer mode → third-party inference (3p)": you can enter a Base URL and Key without logging in. It caught fire in the Chinese-speaking community, with a flood of tutorialsop7418, Gorden_Sun, Tencent Cloud tutorial
2026-09Reports say Anthropic added a model allowlist to the 3p mode, blocking non-Claude modelsTencent Cloud: Claude Desktop quietly blocks third-party models
10-14Codex++ v1.9.0 launches the "Claude Enhancement" page (interface localization, voice input; formerly called "Claude Localization")Codex++ releases
10-22Codex++ v1.11.0 adds Computer Use to Claude Enhanced, so 3p mode works tooSame as above
11-10Codex++ v2.6.0 launches "Claude Desktop local mode," whose "full-feature mode" uses a real profile plus a local CA/proxy to recreate the claude.ai interface. This is ClaudeMom's predecessorv2.6.0
11-12 06:54 UTCCommit: "Claude Desktop full-feature mode core (split from CodexPlusPlus)"; 07:06 creates the repo, and v0.1.0 and v0.1.1 ship the same daycommits
11-12Codex++ v2.6.7 release notes first mention that "Codex++ and ClaudeMom share Claude Code session records"v2.6.7
11-12 night (UTC)A post titled "【开源推广】ClaudeMom …" appears in linux.do's "Resource Collection" section (topic 3166605). The mirror site shows the first 4 replies at 11-13; the linux.do list page shows about 15 replies and about 424 views. The original post is now 404Feidu mirror, linux.do tag page
11-13WeChat Official Account "南京Sol" publishes "No Claude account needed! Turn your API Key into login mode…"Sogou WeChat
11-15Our repo caught its first stars and forks: 22 stars, 0 fork. The first fork landed at 11-15 09:38 UTC (17:38 Beijing time), and forks kept climbing after that. v0.1.3 shipped at 19:37 UTC.repo_day, forks API
11-16 02:56 UTCa Chinese dev forum (linux.sb) reposted the whole thinglinux.sb/topic/32684
11-16 09:12 UTCthe Russian Telegram channel "Вайб-кодинг" (81,446 subscribers) posted: "A totally unhinged project on GitHub, ClaudeMom... it doesn't swap in a fake Claude, it tricks the original client", 21.7K readst.me/vibecoding_tg/4355
11-16aggregator sites picked it up: gotburnout.io published it in 6 languages under the headline "hypnotizing the official client"; AI Qazaqstan in 4 languages; the Telegram channel "Vibe Coding Daily" (3,816 subscribers) translated it into Englishgotburnout, aiqazaqstan, tg.me/gptfresh/1751
11-16 to 11-17公众号接着搬运:「慢慢长大的糖糖」「光谋纪元」「小锋学长生活大爆炸」「每周AI拾光」「泡学堂·GitHub 每日热点速递」「陈嗣俊《免账号解锁 Claude 桌面版的国产项目,在你本机装了张证书》」等,第一页能看到 8 篇search "ClaudeMom" on Sogou WeChat search
11-16 to 11-17It shipped 3 to 4 versions a day (v0.1.4 to v0.1.8). External contributor wuwuzhazha sent repeated Windows/Cowork fixes; Yuanuo opened a PR against upstream DeepSeek support.issues/PR

Growth data

  • Stars and forks (our GitHub repo): 11-15 is 22/0,11-16, 180/28 (+158), 11-17 is 506/102 (+326), 11-18 is 591/122 (+85). The GitHub page showed 11-18 stars and 596 forks at 124. The peak single-day star gain was 11-17, and it fell to 11-18 on 85. That's cooling, not the "still accelerating" the data card claims.
  • Fork timing (forks API, UTC, bucketed every 6 hours): the 11-15 bucket had 37 forks, and the 11-16 00–06 bucket was the largest at 27 (Beijing morning). The day total was 11-16 forks; 73 came after 11-17 (only 13). The earliest forks predate the Russian channel's post, so the first wave was mostly Chinese-speaking.
  • Forks/stars come to about 20%; the median for new repos of similar size is about 11%. Chinese tool repos often look like this: people fork first to keep a copy in case it gets taken down. Some of them did come to submit PRs.
  • Downloads (GitHub Releases, as of 11-18): across 9 versions, the macOS dmg was downloaded 308 times, the Windows exe 692 times, and the auto-update latest.json was fetched 2,082 times. Probably a little over a thousand people actually installed it.
  • We can't get the stargazer breakdown; the GitHub API returned 404, so we can't tell how many stars came from the Chinese community versus the Russian one.

Competitors (doing the same thing)

NameWhereLaunchedScaleHow ClaudeMom differs
Claude Desktop's official 3p mode (Developer Mode → third-party inference)Big-company features2026-04Anyone with the desktop app can use it; dozens of Chinese tutorialsOfficially supported and needs no certificate install. But computer control, voice, and some Cowork entry points are missing, there's no Chinese UI, and reports since September say it restricts non-Claude models.
Claude Cowork via Amazon Bedrock / Microsoft Foundry 3PBig-company features2026EnterpriseGoes through cloud vendors' official channels, aimed at enterprises. Doesn't solve the individual need to use a relay.
farion1231/cc-switch (Claude Desktop panel)Open source2025-08 (Claude Desktop panel about 2026-05)155,198 starOnly writes 3p config, manages providers, and maps models. It doesn't fake a login, so it's as limited as 3p. ClaudeMom goes the other way and supports importing providers from it. They complement each other.
lonr-6/cc-desktop-switchOpen source2026-04-24804 stars, stalledOnly does 3p provider config (DeepSeek, Kimi, GLM, Bailian)
javaht/claude-desktop-zh-cnOpen source2026-04-248,655 stars, +227 in the last 7 daysOnly does a Chinese localization patch, doesn't solve account or feature issues
Jyy1529/claude-desktop_win-zh_cnOpen source2026-04-26678 starOnly localizes the Windows version into Chinese
TonF1917/claude-desktop-multiple-windows-different-apiOpen source2026-10-14985 starMulti-instance: one window uses the official subscription, others go through 3p. Requires an account.
BigPizzaV3/CodexPlusPlus's "Claude full-feature mode"Open source2026-11-10A feature inside a big tool with 33k starsSame author, same core. ClaudeMom is a lightweight standalone spin-off that only handles Claude and doesn't require installing all of Codex++.
S-Trespassing/claw-in-chromeOpen source2026-04-05527 starThe same idea applied to the Claude in Chrome extension: lift login and plan limits, connect custom providers. ClaudeMom does it on the desktop app.
Open-source Cowork alternatives such as openwork / eigent / AionUiOpen source2025 to 202615k to 34k starsRewrites the client from scratch. ClaudeMom runs the original client, so the UI and features match the official ones.

Summary: "Use the Claude desktop app without logging in" was already done by the official 3p mode and cc-switch. A Chinese localization patch with 8.6k stars already existed too. What ClaudeMom adds is faking the login state, which unlocks features that are exclusive to the logged-in version. In open source, ClaudeMom is the first to package this layer as a one-click install with signing and notarization. Its predecessor was a feature the author had put into Codex++ just two days earlier.

Growth tactics (where users come from)

  1. The author's existing users and reputation (verified): Codex++ has 33,192 stars, a Telegram group of 7,422 people, and a QQ group (now at group 4), and it has been written up many times on Zhihu and CSDN (for example "Modding Codex: the enhanced Codex++ is going wild on GitHub!"). The Codex++ v2.6.7 release notes mention ClaudeMom directly. The wave of Chinese forks that started at 11-15 in the afternoon most likely came from these groups or Codex++ users, but we can't see the group chats, so this is unconfirmed.
  2. linux.do open-source promo post (verified, small reach): posted under the community's "open-source promotion" tag, with a LINUX DO friend link added to the README. The post got about 424 views and 15 replies. The original post is now 404, reason unknown.
  3. WeChat Official Account reposts (verified): at least 8 accounts from 11-13 to 11-17, mostly "GitHub hot repos" aggregator accounts. Their titles basically copy the README line "No Claude account needed!"
  4. Chinese AI community reposts (verified): linux.sb and the Xianrenbing community (xianrenbing.com).
  5. Russian Telegram and multilingual aggregators (verified): "Вайб-кодинг" with 81k subscribers and 21.7K reads on a single post, followed by gotburnout (6 languages), AI Qazaqstan (4 languages), and Vibe Coding Daily. This wave hit during the 11-16 to 11-17 window, when daily star gains peaked. There was also a fork renamed reclaw17/Claude-mm with a Russian-language description, which shows the Russian audience was paying attention.
  6. GitHub repo aggregators (verified, low impact): HotRepo, Kapyn, Context7, SkillsLLM, and Nicheloom all list it.
  7. X / big accounts (not found): our x-radar corpus returned nothing for the relevant keywords (even "Claude Desktop" only gave 0 results, so the corpus may be broken). A Brave search of x.com also turned up no posts about ClaudeMom. GitHub trending doesn't list repos less than 60 days old, so it doesn't apply here.

Why it's working

  1. Demand was already primed, and the official product left a gap. When the official 3p mode launched in April, it flooded Chinese social media, and people had already accepted running the Claude desktop app on a relay API key. But 3p lacks computer control, voice, and some Cowork features, and in September it was reported to restrict non-Claude models. Chinese users are afraid of account bans, and tutorials are everywhere. ClaudeMom hits the pain point of people who have a key but no account (or don't dare use one) and still want the full feature set.
  2. The hook is clear in one line. "No Claude account! Turn your API key into login mode," plus a three-column comparison table. WeChat accounts and the Russian channel copied it almost verbatim. The Russian audience was drawn to the technical irony of "tricking the official client" ("hypnotizing Claude").
  3. The author already has a head start. The core code has been running in Codex++, which launched 11-10 and was split out 11-12. So ClaudeMom ships as a finished product on day one: a signed and notarized dmg and exe, auto-updates, and import from cc-switch. The author also brings the trust and community built up around Codex++, which has 33k stars.
  4. Fast fixes, and the community was willing to help. It shipped 9 versions in 6 days, with 4 on 11-16 alone. The author replied to issues one by one, and outside contributors volunteered Windows/Cowork fix PRs.
  5. Cross-language spread. The Chinese community drove the early numbers (forks from 11-15 to 11-16 in the morning), and the Russian Telegram accounts amplified it afterward (11-16 to 11-17 in the afternoon). That's unusual for a single Chinese tool.

Differentiation

  • Differentiator: others (the official 3p mode, cc-switch, cc-desktop-switch) all accept the limits of 3p mode and help you configure 3p. ClaudeMom fakes the "logged in" state and feature flags on your machine, unlocking the logged-in version's features while forwarding conversations to your own upstream. Localization, voice, and computer control are bundled along the way, and all of them already existed in the author's Codex++.
  • Is that why it took off?: Yes. If it only did 3p config, it would just be a stripped-down version of cc-switch. If it only did localization, a patch with 8.6k stars already existed. "No account needed, yet full logged-in features" is the core line of every promo post, and the thing the Russian audience found "insane". Still, the fact that it became a finished product so quickly and got seen right away depended on what Codex++ had built.

Visibility

It's spreading in small circles. Evidence: the linux.do original post got about 424 views and is now 404; the first page of WeChat reposts has 8 articles, mostly from small accounts; and a Russian Telegram channel with 81k subscribers got 21.7K reads on one post, plus the aggregators. We found no English tech press, no big X accounts, and no Hacker News or Reddit threads. It's free and open source, so there's no public revenue to report.

Copycats

  • Renamed or re-skinned forks: reclaw17/Claude-mm (11-16, with descriptions in Chinese, English, and Russian); latentharbor/ClaudeMom (forked, now publishing its own releases, with version numbers already at v0.1.13, ahead of upstream); Xavier189/ClaudeMom (description reads "use Claude locally").
  • We haven't seen an independent rewrite of the same idea yet. Codex++ itself was renamed and redistributed in the past (bighamx/CodexSharp, with the sponsor slot removed), and ClaudeMom will likely see the same.

What others can borrow

  • Split the most viral feature out of a big tool into its own repo. The feature was already proven in the older product, and as a standalone project the name is easier to remember and the description can be a single hook line. A big tool's README and group chats can't make the feature the headline, but a standalone repo can.
  • Look for the gap between the official "downgraded mode" and "full mode". When the vendor offers a limited login-free or third-party mode, the community will want "no login, full features." The same approach has already appeared in Claude in Chrome (claw-in-chrome) and in Codex (Codex++'s API-mode unlock plugin), and it can carry over to any AI client with a gap between the official logged-in version and the API version.
  • Use a three-column comparison table in the README to show where you stand, plus one line that only your target users will get ("✅ but don't dare turn it on").
  • Follow community rules when posting (the linux.do open-source promo tag, friend links), and productize releases (signing and notarization, auto-update, one-click import of configs from popular tools) to lower the installation barrier as much as possible.

Still unclear

  • The direct source of the Chinese fork wave at 11-15 09:38 UTC: possibly Codex++'s QQ and Telegram groups, possibly a WeChat article or an X post. Group chats and article read counts aren't visible, so this is unverified.
  • Why the linux.do original post is 404 (deleted by the author, moved to a login-only section, or removed for a rule violation) is unverified.
  • How many stars came from the Chinese community versus the Russian one: the stargazer API isn't open, so we can't split them.
  • Whether Anthropic will act against it (client update blocks, DMCA takedowns, bans on accounts linked to it), and how long its stars and usage can hold up. Daily star gains have already fallen back to 85 from the peak of 11-18.
  • Real usage: about 1,000 downloads of the installers and about 2,000 auto-update checks. We can only roughly infer an install base in the thousands.

This article was written by AI from public data and the data we collect daily. Numbers are sourced, and inferences are labeled.