
"No need to buy the Mind Ball": one reply took an Android open-source app from 25 to 230 stars in two days
What it is An open-source Android app that keeps the phone listening all day in the background, recording only when someone speaks, transcribing on-device, identifying who is speaking, then having a large model organize it into to-dos and memories, with no hardware to buy and no account to register.
Why it took off OPPO Mind Ball costs ¥499, requires an invite code, and was being recommended across the Chinese-language circle; the author replied under a recommendation post by Xiangyang Qiaomu, "No need to buy the Mind Ball, same effect." Qiaomu replied "Has something to it," and the next day goDavi included it in a weekend recommendation. The app itself had shipped over twenty versions in two weeks and was usable right after install, which caught the wave of people.
CHow you could follow it Don't build another TaoSheng; copy its playbook: find an AI hardware gadget that is being recommended and needs buying, waiting, or cloud access, offer a pure-software substitute using your phone or existing earbuds, and reply under a big account's recommendation with a line like "No need to buy X"; or take the data TaoSheng surfaced and build the knowledge-organization piece it didn't.
- Data (updated daily) · Repo created 2026-10-24 · 433 stars · 7 days +65
GitHub · stars Data 2026-10-31 → 2026-11-20 (21 days, 21 observations). Points before 10-28 come from deeply / archive backfill, not every day has one
TaoSheng is an open-source Android app that turns the phone into an always-on recording pendant: it records only when someone speaks, transcribes on-device, identifies who is talking, and has a large model organize it into to-dos and memories. It had only 25 stars two weeks after its repo was created, until the author replied "No need to buy the Mind Ball" under a big account's post about OPPO Mind Ball being worth recommending. The move is one anyone can copy, but the TaoSheng concept has already been executed thoroughly by someone else, and in two days it gained only about two hundred stars, so the fame it could capture is limited.
A reply from Nov 7
On the evening of Nov 7, 2026, at 6:30 p.m. Beijing time, the big Chinese AI-circle account Xiangyang Qiaomu (@vista8) posted a casual remark: a friend's OPPO foldable unfolds to a 181-degree angle, "dead flat"; at the end he added offhandedly, "Lately OPPO's product reputation keeps improving, like the AI Mind Ball, 24 hours of recording logged, a bit of a recommendation." That post later got 119k views.
Less than six hours later, an account called 0xCoach (Atao) replied underneath:
Bro, try my approach. No need to buy the Mind Ball, but it achieves the same effect, more flexible and easier to use.
A GitHub link followed. 4 minutes later Qiaomu replied to him: "Wow, has something to it, I'll try it on my Android phone someday." That reply got 19.4k views, and Atao's post 27.9k. Some people below said "it runs on existing Android phones, more practical than buying yet another new gadget," while others poured cold water: "The problem with using a phone is that most of the time it's in your pants pocket, so the recording quality is poor."
At noon the next day, another Chinese AI blogger, goDavi (@corndogjpn2), posted a weekend recommendation putting TaoSheng alongside two other projects: the Arion voice input method and Cicada meeting minutes. His verdict: "Something a master wrote, and although it's open source and free, in features, maturity, and positioning it has already caught up with the commercial paid apps it competes with."
GitHub stars (recorded once a day):
| Date | Stars | Added that day |
|---|---|---|
| 10-31 | 21 | — |
| 11-07 | 25 | Only 4 more in a week |
| 11-08 | 65 | +40 after Qiaomu's reply |
| 11-09 | 230 | +165 after goDavi's recommendation |
By 11-09 it was at 244 stars and 28 forks. The fork-to-star ratio is normal, not looking artificially inflated.
What the Mind Ball is, and why that line worked
The OPPO AI Mind Ball is a 9.8-gram clip-on device launched on Sep 17 and sold from 22, offering 24 hours of recording, voiceprint speaker identification, and AI organization, at an early-bird price of ¥499, invite code required to buy. Many people in the Chinese-language circle were talking about it then, and some said it "outclassed every recording product of the past two years." There was plenty of interest, but not everyone was willing to spend ¥499 and wait for an invite code, nor to hand their recordings over to OPPO's cloud.
TaoSheng's README has a section titled "You and the AI recording pendant are only one string apart":
Look at the AI recording cards and recording pendants on the market, and open them up: they're just a Bluetooth microphone plus a battery. ... Hang a Bluetooth earbud around your neck on a string, and you've got an AI recording pendant.
The accompanying image is simply a pair of Bluetooth earbuds hanging from a string. This line strips away the hardware and is understood at a glance, which makes it well suited for replying under hardware-hype threads.
Once you click in, can the product hold up?
A good line alone isn't enough; people who click in need to install it and use it right away. TaoSheng built its repo in Oct 24 and released its first public version in Oct 26; by Nov 8 it was at v0.2.19, with over twenty versions in two weeks and 29 features listed in the README. The key ones:
- Zero hardware: the phone's microphone is enough; it also works with Bluetooth earbuds and clip-on mics, and switches back to the phone automatically when they disconnect.
- Local first: on-device voice activity detection, no recording during silence; the author's measured battery drain is about 0.5% per hour; you can choose a local model for offline transcription, so audio never leaves the phone; no account registration, and the author runs no server; original recordings are automatically deleted after 3 days by default.
- Speaker identification and to-dos: on-device voiceprints recognize "me" and people you know; commitments made in the original speech automatically become to-dos, which go off like an alarm when due.
- Full integration with the Chinese ecosystem: DeepSeek, Doubao, Tongyi, Kimi, and more than ten other models switchable with one click, with automatic fallback to a backup model when the primary goes down; push notifications connected to WeCom, Feishu, DingTalk, Server 酱, and 10 other channels; backup to Jianguoyun; installation packages mirrored in China on ModelScope.
- Direct APK download, with a 1-minute 45-second promo video on the homepage.
So before it blew up on X, it already had a small group of real users: someone submitted a fix the same day as the first public release, resolving crashes on the OPPO Find N6; external PRs followed, adding Android 9 compatibility and voiceprint sample management, which the author merged and credited in the README. Across the GitHub release packages, downloads total about 850 (11-08); in-app update downloads via delta packages and the ModelScope mirror aren't counted in that figure.
One user's review after a day of use was reposted by the author: "It meets or exceeds nearly all my expectations. By contrast, the memory bracelet I bought for 500 has some problems, like not letting you listen to the recordings... too much black box."
Another point that makes the vibe-coding crowd willing to share it: the README states that nearly all the code was written by AI, with Claude handling architecture and UI, and Codex handling the low-level pieces like recording, transcription, and the database; nearly a hundred task cards given to the AI are all public in the repo.
Who else is doing the same thing
| Product | What it is | Scale | Compared with TaoSheng |
|---|---|---|---|
| OPPO AI Mind Ball | Clip-on hardware + OPPO cloud | ¥499, bought with an invite code | Needs hardware, data goes to the cloud, no original audio kept |
| Anker AI Recording Beans (co-branded with Feishu) | Hardware + Feishu | ¥899 | Meeting-focused, tied to Feishu |
| Plaud, Bee, Limitless, etc. | AI recording cards / pendants | Plaud Android app ~820k installs | All require buying hardware and a transcription subscription |
| Omi | Open-source hardware pendant + app | ~13.9k stars on GitHub | Can also run on a phone alone, but the English ecosystem needs an account and uses its cloud; TaoSheng itself says it was inspired by it |
| ShengWuji (shengwuji) | Open-source Android app | 143 stars | Also uses offline local recognition, but focused on quick-note capture and object management |
| Cicada (zhile) | Open-source meeting minutes / live captions | 1,888 stars (11-09). | Recommended by the same people the same weekend, but it covers meetings, not all-day recording |
Searching GitHub for 'Xinlicheng' (heart-force ball), 'all-day recording', and 'recording pendant' turns up only this one solid project. The 20-plus forks are all at 0 stars, mostly people opening PRs. So the position of 'turn your phone into an all-day recording pendant for Chinese users' currently belongs only to TaoSheng, and it's already quite complete.
Its weak spots are clear: it only supports Android 10+ phones with 64 bits; iOS doesn't let apps keep the mic running in the background, so that's a no-go; a phone in your pocket picks up audio poorly; and all-day recording of other people's speech carries legal and social risk, which the README itself warns about.
How many people know about it now
Very little. The only traction I could find is the reply from 乔木, goDavi's weekend recommendation, and a few reposts by the author. No press coverage, and nothing on Hacker News, V2EX, or Xiaohongshu. The 244 stars two days in are a number from a small Chinese corner of X.
If it were you
Bottom line: Stop building another TaoSheng. Copy how it got its stars instead: when some AI hardware is being talked about, build a pure-software substitute and reply under the big accounts' seeding posts with 'No need to buy X, same effect.' That move costs nothing and needs no celebrity, so anyone can use it. But this topic is already owned by TaoSheng, which is feature-complete and shipped twenty-plus versions in two weeks, so latecomers will struggle to beat it. And even at that level of execution, it only reached a few hundred stars.
Who it's for: People who can write Android or desktop code and can use Claude Code or Codex to ship an installable, working app in two weeks; people who follow the Chinese AI scene closely and know which hardware is being hyped.
Who it's not for: Anyone trying to build yet another all-day recording app for Chinese Android; anyone who wants thousands of stars or wants open source to build their name, since this kind of topic currently tops out at a few hundred stars; anyone who only has an iPhone and can't keep the mic running in the background.
How to do it:
- Pick a different hardware category. Watch AI gadgets that are being hyped, need to be bought, need waiting, and depend on the cloud: recording cards, a single feature of AI glasses, desktop AI ornaments, AI toys for kids. Ask yourself: once you take it apart, is it just 'mic/camera + battery + one app'? If so, build a software version on a phone or with off-the-shelf earphones.
- Or build the part TaoSheng hasn't: catch its users and do the 'all-day recording → personal knowledge base' stage, for example automatically organizing each day's memories into Obsidian or Notion, or building a desktop tool for organizing and searching. TaoSheng can already export verbatim speech to Markdown and JSON, so you don't need to compete with it on this stage.
- Whichever you choose, the first screen of the README needs one line anyone can understand plus a picture, matching TaoSheng's 'one string' idea. Make the installer directly downloadable, add a domestic mirror, and include a one-to-two-minute demo video.
- Release frequently in the first two weeks, reply to issues the same day, merge external PRs the same day and credit them in the README, and let the first few dozen users help you find device-specific bugs.
How to get users: On Chinese X, find big-account posts discussing that hardware and reply with your link. Keep it short and make a direct comparison ('No need to buy X, same effect'). Don't post long threads. If a big account replies, that's your first wave. Then find bloggers who publish weekly 'weekend picks' or 'open-source picks' and get them to feature you alongside other good projects that week. Showing how the project was built with AI (public task cards) is itself a shareable hook in the vibe-coding crowd.
When to move: In the weeks after the hardware launches, while it still needs invite codes or hasn't rolled out widely. TaoSheng's post came about a month and a half after the heart-force ball went on sale, while interest was still high. Once the hardware is easy to buy and reviews are out, 'no need to buy' loses its pull.
Signs you misread it: You post the reply, the big account responds, and after three days you still haven't reached 100 stars. That means the big account only mentioned the hardware in passing and no crowd actually wants it. Or you find an open-source substitute for that hardware already on GitHub (like Omi in the English world), which means the position is taken. Then switch hardware or switch audiences.
Research notes: competitor table, acquisition evidence, full research report
- Competitors
Name Where Launched Scale How it differs from this one OPPO AI Mind Ball Other 2026-09-17 launch / 09-22 on sale Early-adopter price ¥499, buy with an invite code; strong word of mouth on Chinese X Requires a clip-on mic, uses OPPO's private cloud, keeps no original audio; TaoSheng needs no hardware, keeps data on-device, and lets you choose the model Anker AI Recording Beans (co-branded with Feishu) Other 2026-01-19 ¥899, available on JD Hardware plus the Feishu ecosystem, meeting-focused; TaoSheng targets individual all-day lifelogging and is free Omi(BasedHardware/omi) Open source 2024-03 13,924 stars; the app has a phone-mic-only mode English ecosystem, requires an account and cloud; TaoSheng has no accounts, does offline Chinese transcription with SenseVoice, and uses domestic models and push channels Plaud / Bee / Limitless and other AI recording cards and pendants Other From ¥2023 Plaud Android ~820k installs (Google Play library) All require buying hardware and a subscription; TaoSheng's pitch is 'pendant = Bluetooth mic + battery', and any Bluetooth earphone on a lanyard works iFlytek / Honor KUMI / WeChat Input AI recording pens, etc. Big-company features various Big Mainly meeting transcription; no all-day background capture, voiceprint, or to-do memory wquguru/cicada (Cicada) Open source 2026-11-01 1,888 stars (11-09). Recommended by the same people the same weekend; does meeting captions, live interpretation, and minutes, not all-day capture fantasyao/shengwuji (ShengWuji) Open source 2026-07-05 143 stars Also uses SenseVoice offline recognition, but offers quick-note capsules and object management, not all-day recording kiyoakii/xinliqiu-sdk Open source 2026-10-24 29 stars Unofficial SDK for heart-force ball; serves heart-force ball users rather than replacing it adamcohenhillel/ADeus, memex-lab/memex Open source 2024-01 / 2026-03 3,437 / 829 stars ADeus is open-source wearable hardware; memex is an AI journal with manual entries - Where users come from
- Verified X reply to a big account's hot post (riding the heart-force ball hype): 11-07 16:09 UTC @0xCoach replied under @vista8's 'AI heart-force ball is kind of a hype' post with 'No need to buy the heart-force ball, but you get the same effect' plus the repo link; @vista8 replied 4 minutes later, 'Wow, something to it, I'll try it on my Android phone sometime', 19.4k views; the next day the repo got 25 → 65 stars
- Verified Second wave from the X weekend-picks post: 11-08 12:31 UTC @corndogjpn2 (GoDavi) recommended TaoSheng alongside Arion and Cicada, saying its completeness matches paid commercial apps; stars gained that day: +11-09165
- Verified Listed on AI Coding Hunt (Xiangyang Qiao Mu's own site): 11-06 10:19 listed 'Taosheng' when it had 18 stars, 11 site views, and 0 likes, with almost no traffic
- Verified Author's X account and user reputation: @0xCoach's pinned user praise reads 'Used it for a day… beats almost everything I expected; 500 I bought memory bracelets with too many black boxes'
- Inferred Seed users before the spike: 10-26 On day one of the first release, an external PR already fixed an OPPO Find N6 crash; 10-27~29 feature issues followed; the source (the author's X or WeChat groups) could not be confirmed
- Inferred media / HN / V2EX / linux.do / Xiaohongshu / ads: no search hits; x-radar corpus 0 entries
0xCoach/TaoSheng (TaoSheng): Turn an Android phone into an AI recording pendant
Research date 2026-11-08/09. Repository github.com/0xCoach/TaoSheng, Kotlin / Jetpack Compose; the README says MIT (GitHub detects NOASSERTION). Author 0xCoach (X @0xCoach, 'Ah Tao', bio: 'sharing AI tools, vibe coding, agents, and open-source projects'). GitHub account registered 2016 years ago, only this one public repo, 3 followers.
Bottom line
Tao Sheng is a pure-software, free, local-first Android alternative to the OPPO AI Mind Ball (launched 2026-09, priced at 499 yuan, invite-code only). It does the same thing: records all day, recognizes who's speaking, then uses AI to pull out to-dos and memories. The direct trigger was very simple. The author 11-07 replied under a post by Chinese AI account 向阳乔木 (@vista8) saying 'No need to buy the Mind Ball, you can get the same effect.' Vista replied, 'Oh, interesting, I'll try it on my Android phone.' That reply got 19.4k views. Then goDavi and others shared it in their weekend recommendations, and stars went from 25 to 230 in two days. The project could absorb the traffic because the product was already solid: 25 versions in two weeks, 29 features, offline transcription, speaker recognition, and a direct APK install, not an empty shell. So far it's only circulating in small circles on X's Chinese AI scene. No media coverage, and no copycat repos yet.
1. Data
- Star curve (our github repo_day): 10-24 repo created; 10-31 21 stars, reaching 11-07 with only 25 stars (+4 in one week); 11-08 65 (+40), 11-09 230 (+165); forks 6 → 26. 11-08 current figures pulled live from the GitHub API: 243 stars, 27 forks. The repo is 16 days old, under 60 days, so it can't make trending yet.
- Repo status: 'newly emerged', gain_7d 209, no suspect flag; fork/star ratio ≈ 11%, within the normal range, not like star-farming (the stargazer timestamp endpoint returned 404/401 with our token, so no per-user times were available).
- APK downloads (download counts from GitHub release assets, 11-08 check): about 900 total across all versions; most downloaded are v0.2.17 (135) and v0.2.12 (89). There is also a domestic mirror on ModelScope (modelscope.cn/OpenRecode/TaoSheng-releases), but its download counts are not visible.
- Release cadence: 10-26 v0.2.0 "first public release" to 11-08 v0.2.19, with a string of preview builds in between. About 25 releases in two weeks; roughly 432 commits.
- Community: 13 issues/PRs. The first one was 10-26 (SIGILL crash on the OPPO Find N6, PR fix from outside user ethansunqing). From 11-05 on, external PRs kept coming in (Android 9 compatibility, voiceprint sample management, custom transcription endpoint, multi-select merge). The author merged them all and credited contributors in the README. This means a small group of real Android users was installing it and reporting device issues before the X buzz.
2. Competitors: who else does the same thing
| Product | Where | Launched | Scale | How it differs from TaoSheng |
|---|---|---|---|---|
| OPPO AI Mind Ball | Hardware + OPPO phone | Launched at the 2026-09-17 developer conference, on sale 09-22 | Early-bird price ¥499, purchased with an invite code (IT Home, Phoenix, Sina). On X there's a word-of-mouth post saying it "beats every recording product of the past two years". | Requires buying 9.8g clip-on hardware, runs through OPPO's private cloud, and "keeps only text, not the original audio". TaoSheng needs no hardware, data stays on the device, original audio can be kept, and you choose the model. |
| Anker AI Recording Beans (co-branded with Feishu) | Hardware + Feishu | 2026-01-19 | ¥899, sold on JD | Tied to the Feishu ecosystem and aimed at meetings. TaoSheng targets individuals doing all-day lifelogging. |
| Omi(BasedHardware/omi) | Open source + hardware pendant + iOS/Android app | 2024-03 | GitHub 13,924 stars (our repo library). The app has a "phone microphone only" mode (help.omi.me). | TaoSheng's README says it was inspired by Omi. Omi also works phone-only, but it's English-first, requires an account, and goes through its cloud. TaoSheng has no accounts, offline transcription with local SenseVoice, domestic models (Doubao/DeepSeek/Qwen/Kimi), and domestic push channels (WeCom/Feishu/DingTalk/Server Chan/Jianguoyun). |
| AI recording cards / pendants like Plaud, Bee, Limitless | Hardware + app | From ¥2023 | Plaud's Android app has about 820k installs (our Google Play library) | All require buying hardware and paying for transcription subscriptions. TaoSheng's pitch is "the recording pendant is a Bluetooth mic + battery": hang any Bluetooth earbud around your neck and it's a pendant. |
| iFlytek, Honor KUMI, WeChat Input AI recorder, DingTalk, etc. | Big-company features | various | Big | Mainly meeting transcription; no all-day background capture, voiceprint, or to-do memory |
| kiyoakii/xinliqiu-sdk | Open source | 2026-10-24 | 29 stars | An unofficial SDK for the Xinliqiu (Mind Ball) pendant, meant for its owners, not a substitute. |
| fantasyao/shengwuji (ShengWuji) | Open-source Android | 2026-07-05 | 143 stars | Also uses sherpa-onnx SenseVoice for offline recognition, but it does "flash capsule notes + item management", not all-day capture. |
| ADeus, memex-lab/memex, etc. | Open source | 2024-01 / 2026-03 | 3,437 / 829 stars | ADeus is an open-source wearable hardware project. memex is a manually logged AI diary, not continuous recording. |
| wquguru/cicada (Cicada) | Open source | 2026-11-01 | 1,888 stars (11-09). | Open-source meeting-notes / simultaneous-interpretation tools recommended by the same people the same weekend, with 8× TaoSheng's stars. They cover meetings, not all-day capture. |
Conclusion: I could not find a second open-source Chinese implementation of "use the phone itself as an all-day recording pendant" on GitHub (searching for all-day recording, Xinliqiu, and Android lifelog over the past month turns up only this one). On the commercial side, most competitors are hardware. The closest is Omi's phone mode, but it doesn't serve Chinese users or domestic models.
3. Where users came from (growth tactics)
- A big account replying to a trending post (main driver, confirmed). On 11-07 10:34 UTC, Xiangyang Qiaomu (@vista8, a former ByteDance AI product manager and a major Chinese AI-circle account with 10k+ posts on X) posted about OPPO's foldable, mentioning in passing that "OPPO's product reputation keeps improving lately, like the AI Mind Ball, 24 hours of recording, kind of a recommendation." At 16:09 UTC 0xCoach replied: "Bro, try my solution. No need to buy the Mind Ball, but it gets the same effect and it's more flexible and useful", with the repo link. 4 minutes later Qiaomu replied: "Wow, there's something to this. I'll try it on my Android phone another day." That reply got 19.4k views. Someone below commented "it runs on the Android you already have, more practical than buying another new toy", while another questioned "the phone is in your pocket most of the time, so the audio pickup will be poor." Stars then went from 25 to 65.
- Sources: x.com/0xCoach/status/2119099904350171374, x.com/vista8/status/2119015563670532515, x.com/vista8/status/2119101016021025251.
- Weekend roundup post, second wave (confirmed). On 11-08 12:31 UTC, goDavi (@corndogjpn2, "Dog V International") posted "Played with several very high-quality projects this weekend", listing TaoSheng alongside @Stv_Lynn's voice-input method Arion and @wquguru's Zhiliao. Comments: "Apps built by veteran coders... maturity and polish are high" and "though open source and free, it's caught up with paid commercial apps." 11-09 single-day +165 matches this timing. Also, an unboxing post for a recording card by @ericoding99 surfaced a TaoSheng link in search results (whether it was a comment or a quote is unverified). The author's pinned X post is a user review: "Used it for a day... met or exceeded nearly every expectation I had. By contrast, the 500 memory bracelet I bought... too much black box."
- Included in Qiaomu's own AI Coding Hunt (confirmed, but it had almost no traffic at the time). aicodinghunt.qiaomu.ai listed "Taosheng" (misspelled) on 11-06 10:19, when the GitHub project had 18 stars, 11 site views, and 0 community likes. So the project had already appeared in Qiaomu's circle once before the author replied under his post.
- Small early-user group before the breakout (existence confirmed, source unverified). 10-26 On the first public release day, an outside user already submitted a PR fixing a crash on the OPPO Find N6. From 10-27 to 29 feature issues trickled in, and one v0.2.12 build got 89 downloads. Where these people came from can't be traced (the author's X, WeChat groups, or Moments could all be sources).
- Channels we didn't see: no press coverage, no HN / V2EX / linux.do / Xiaohongshu / Jike posts (search came up empty); our x-radar corpus returned 0 hits for 「涛声」 and 「心力球」 (the corpus doesn't cover them well); no ads (open source and free, so there's no paid product to advertise).
4. Why it took off
- Timing: OPPO's Xinliqiu launched in September with limited invite-code sales, which turned "all-day recording + AI memory" into a hot topic in the Chinese scene. There's a group of people who want it but can't buy it, or who don't want to pay ¥499 for hardware plus a cloud black box. TaoSheng's one-liner "no need to buy the Mind Ball" speaks directly to that demand.
- The narrative is easy to repeat: "You and an AI recording pendant are only separated by a cord. Take the pendant apart and it's a Bluetooth mic plus a battery," paired with a picture of a Bluetooth earbud hanging from a string. This "hardware demystified" framing fits naturally as a reply under hardware hype posts.
- The product holds up: it's not a demo, it's an APK you can install and use. 29 features: on-device Silero VAD that records only human speech, about 0.5% battery drain per hour; SenseVoice offline transcription on Snapdragon NPU; CAM++ speaker recognition; strong reminders for to-dos; daily review; 10 domestic push channels; Jianguoyun backup; over a dozen domestic models switchable with one tap, with automatic fallback. It also guides users through keeping the app alive on Xiaomi and ColorOS. goDavi and the user the author pinned both praise its "polish".
- Privacy and cost selling points: no account, no server, original audio auto-deleted after 3 days by default. Local transcription is free, and organizing content uses your own DeepSeek key, so costs are very low. Compared with the Xinliqiu's "goes through OPPO's cloud" and recording cards' "subscription required", that's a reason users will try it.
- "Written by AI" is itself a topic: the README has a dedicated section saying nearly all code was written by Claude (architecture and UI) and Codex (low-level), with close to a hundred task cards publicly posted in docs/任务卡/. That hits right at the interest of Chinese vibe-coding circles on X (Qiaomu and goDavi are both part of it).
- Fast author responsiveness: 25 releases in two weeks, external PRs merged within days with public credit, so 2 weeks in there were 5 outside contributors.
5. Differentiation
What really sets TaoSheng apart from similar tools is the combination of zero hardware, local-first, and a domestic ecosystem:
- Versus hardware (Xinliqiu, Anker AI recording beans, Plaud, Omi pendant): nothing to buy, your phone or any Bluetooth earbud or clip mic serves as the audio source.
- Versus Omi's phone mode: no account, no cloud, offline Chinese transcription, domestic models and push channels.
- Versus open-source peers: it's the only one that turns all-day background capture + voiceprints + LLM organization into to-dos and memory in a complete Android app.
This difference is why it took off, but it works on two levels: what gets people to click is the "no need to buy the Mind Ball" positioning (distribution via riding a trend), while what makes people star and retweet it is the polish. The drawbacks are clear too: it only supports arm64 Android 10+, and can't run on iOS (the system doesn't allow always-on background mic access); pickup is poor with the phone in your pocket (already pointed out in replies); and recording others all day carries legal and social risks (the README includes its own disclaimer).
6. How many people can see it
It's spreading in small circles. Evidence on X: the traceable spread is Qiaomu's reply with 19.4k views, goDavi's weekend roundup, and a pinned user review from the author; stars are 230 to 243. No media, no HN/Product Hunt, not on trending (the repo is under 60 days old, so it couldn't have made it anyway). It's free and open source, so there's no revenue.
7. Copycats
None yet. Searching GitHub for "Xinliqiu", "all-day recording", and "recording pendant" only returns this project. The 27 forks are personal forks with 0 stars each (mostly used for opening PRs).
8. What to borrow
- Replying to a big account under competitors' hype posts (especially big-company hardware) with "no need to buy X, same effect" is a zero-cost, high-conversion cold start, as long as the link leads to a product that actually works (direct APK download + a 90-second demo video + a domestic mirror).
- The "hardware demystified" narrative: break a hot hardware category into "mic + battery + software," build only the software, and offer an instantly understandable image like "a Bluetooth earbud on a string."
- An open-source app aimed at domestic users with domestic models, domestic push services, Jianguoyun, and ModelScope mirrors all wired up, which lowers the bar well below overseas peers.
- "AI-written app + public task cards" is itself shareable content within the vibe-coding crowd.
- Possible gaps: Xinliqiu and recording-bean users want things that aren't black boxes, such as seeing the original audio, adjusting when transcription runs, and swapping models. iOS can't do all-day background capture, but a combo of "Bluetooth clip mic + phone" could work. A paid hosted version and desktop organization or Obsidian sync are currently unclaimed.
9. Still unclear
- Among the stars between 11-08 and 09, how many came from Qiaomu's reply versus goDavi's post (stargazer timestamps are unavailable). Whether bigger accounts reshared it two days later.
- Where the early-user group (people behind the issues/PRs from 10-26 on) came from.
- Real usage: about 900 GitHub APK downloads. Volume from the ModelScope mirror and in-app incremental updates isn't visible.
- The author 0xCoach's X follower count, and whether a paid version or iOS version is planned (the roadmap mentions an iOS version).
- Xiangyang Qiaomu's follower count is unverified (I only saw 10k+ posts and a reply with 19.4k views).
Sources
- Repo README, releases, issues, forks: github.com/0xCoach/TaoSheng (GitHub API, checked 2026-11-08)
- Star curve: our github library repo_day / repo_profile (2026-10-31 to 11-09)
- X: x.com/0xCoach/status/2119099904350171374; x.com/vista8/status/2119015563670532515; x.com/vista8/status/2119101016021025251; x.com/corndogjpn2/status/2119407409647292672; x.com/ericoding99/status/2119019245342482781; x.com/0xCoach (pinned); x.com/JeffreyCalm/status/2111391479797301707
- AI Coding Hunt: aicodinghunt.qiaomu.ai/p/淘生-taosheng/
- OPPO Xinliqiu: ithome.com/1/005/942.htm, feng.ifeng.com/c/8xGw38vaiYy, finance.sina.com.cn/tech/discovery/2026-09-22/doc-inistmex5631414.shtml
- Anker AI recording beans: finance.sina.com.cn/tech/2026-01-19/doc-inhhvfsc9940714.shtml
- Omi phone mode: help.omi.me/en/articles/13153576-using-omi-with-phone-only; BasedHardware/omi stars from our github library
- Plaud install count: our Google Play library app_profile
- Similar open-source repos (cicada, shengwuji, xinliqiu-sdk, ADeus, memex): our github library