Everyone else is building a Hongguo client; he built an SDK layer: how openhongguo got 329 stars in 11 days

Hongguo Short Drama is a free short-drama app under Douyin. This year, in 9 and October, a dozen-plus GitHub repos popped up called "Hongguo desktop" and "Hongguo TV"; the biggest has 5,400 stars. In Oct 26, a developer who had built a temp-email project with 2,800 stars didn't make yet another client. He made an SDK anyone can import. 11 days later it had 329 stars and 93 forks. The biggest star jump came from a plugin that plays short dramas in the sidebar while an Agent works. This piece covers how it found an opening in a crowded hot trend, and what you can take from it if you want your open-source project to get stars.

"Who could resist catching an episode of Hongguo between pedal strokes?"

On the morning of Nov 5, frostming, the author of PDM, posted on X: "I declare DeepSeekHarness is way more badass than pi. Who could resist catching an episode of Hongguo between pedal strokes? Serious cc @yihong0618", with a screenshot of a drama playing in DSH's right sidebar. By 11-06 the post had 120k views, and someone replied "suggest shipping boss-detection mode."

His plugin is called dsh-hongguo: while an Agent runs a task, a prompt appears above the input box; click a title and it starts playing, and it pauses automatically when the Agent stops. The second line of the plugin README reads "data from openhongguo," and installing the plugin pulls openhongguo in via npm.

Someone built a webOS version for LG TVs the same day (that post got 6,000 views). openhongguo's daily npm downloads went from a typical few dozen to 220 after 11-04, then 11-05 to 309 (npm download stats); GitHub stars jumped by 63 in one day (11-06).

What it is

An unofficial Go SDK for Hongguo Short Drama, Apache-2.0 licensed. It sends requests as an anonymous device with no login, and can browse channels, search dramas, read trending and ranking lists, fetch episode lists, and get playback URLs and content keys for each resolution.

The selling point is "one core, six languages": every version beyond Go is compiled from the same Go code. Node.js uses WebAssembly, Android and iOS use gomobile, and there are React Native and Flutter packages, published to npm, Maven Central, SPM, and pub.dev. In Node, five lines of code take you from search to a playback URL. The repo also ships a CLI, ohg, where ohg play <title> calls mpv directly to play.

The author is beilunyang. His earlier project moemail (temp email) has 2,836 stars and 2,604 forks, so a group of developers already knew him.

How well it's growing

Source: GitHub daily records, npm download stats. Over the last 7 days +257. Forks 93, fork/star ratio about 28%, versus a median of about 11% for new repos over the same period. Many of the people coming in are actually using it to write code, not just starring it.

Where the stars came from

First, the author posted on linux.do (10-27). The title states the use case directly: "Help you build your own Hongguo app." The body lists what it can do, then gives install commands for six languages and a Node example, and ends with "If you use it to build your own Hongguo short-drama app, or come up with something fun, share it in the thread." It got 59 likes and dozens of replies that day. That invitation was later fulfilled: 11-04 someone posted on linux.do saying "used the openhongguo SDK to solve 90%" and built a web big-screen player.

Second, Ruan Yifeng's weekly (10-30). The author submitted to the weekly's repo himself, and it landed as item 13 in issue 417. The weekly publishes on Fridays, and stars jumped from 72 to 148 the next day. Sites like zendot and bytesizego (which wrote a Go code walkthrough) also reposted it the same day, so how much of the jump came from the weekly alone can only be inferred from timing.

Third, downstream work drove traffic (11-05). That's the post from the opening. The DSH plugin was hot at the time, and "slacking off while an Agent works" was itself a meme people would repost, while the plugin README pointed its data source back to openhongguo. The +63 that followed 11-06 lines up with this, though again the link is inferred from timing.

No sign of ads, paid acquisition or star-buying. It all stays inside the Chinese developer circle: linux.do, Ruan Yifeng's weekly, the frostming and yihong0618 crowd on X, a few Telegram channels. No media coverage.

Why this one and not the dozen clients

Third-party Hongguo projects were already crowded:

They all make finished products for people who watch. 9, and the October wave, showed both that many people want to watch Hongguo on PC and TV, and that many developers want to build their own version. But those who wanted to build had to either take apart someone else's client or reverse-engineer from scratch. openhongguo filled that gap, and to date no second Hongguo SDK exists.

It's also easy to pick up and use because the engineering looks like a proper library: packages are published to every package manager, error codes, pagination dedup, and automatic device rotation are all wrapped, and CI includes API contract checks. Developers now have AI help, so the remaining 10% of the UI can be written in a few days; that's how dsh-hongguo and the webOS version came about. Every time someone makes something fun with it, they promote it for free.

The cost is the ceiling: people who watch dramas don't star an SDK, so its stars sit an order of magnitude below finished clients.

Risks

It reverse-engineers ByteDance's endpoints and handles content keys for decryption. The README has a long disclaimer ("no commercial use," "no large-scale scraping"), but endpoints can change at any time, and it could receive a DMCA notice. Nobody can say when it might stop working. Anyone following this path should think that through first.

If it were you

The verdict: "Build an SDK one layer below a crowded hot trend" is a playbook you can copy, but Hongguo's slot is already taken. It's worth doing, with a low star ceiling and legal risk to carry.

Best for: developers who can reverse-engineer and write cross-platform libraries, and who have some presence on linux.do, V2EX, or the Chinese-language corner of X, and can post and submit their own work.

Bad fit for: anyone aiming for thousands of stars (only finished clients reach that level), and anyone unwilling to take on copyright and platform-API risk.

How to do it:

• Look for the next hot app where third-party clients have already piled up but no reusable library exists. The bar: five or more clients or downloaders for the same app appearing on GitHub. Then build the SDK, not another client.
• If you still want to ride Hongguo, do what it hasn't done yet: wrap openhongguo as an MCP server or agent tool, make a Python version, or just build the most fun downstream app. The stars will go to you.
• Ship packages the same way: one core, compiled to multiple platforms with gomobile / wasm, published to every package manager, with five runnable lines of example on the first screen of the README.

How to get users: Follow its sequence. Post an open-source promo on linux.do titled "Help you build your own X," and end by inviting people to share their work → submit to Ruan Yifeng's weekly before Friday → either yourself or with a well-known developer, get a downstream project that makes a screenshot-worthy, repost-worthy meme (a plugin, a TV version), and state in the README "data from xxx."

When to move: Third-party waves around a hot app usually last one or two months. Ship after clients have piled up but before the platform starts blocking the endpoints. For Hongguo, the window is roughly now.

Signs you've misread it: Two weeks after launch, fork/star still hovers around 10%, meaning nobody is really building with it, and developers wanted a finished product instead. Or the target app's endpoints change several times a month and downstream projects keep breaking, so the SDK layer can't hold.