The commit at 4 a.m. the day after the platform launch: how dsh-im became DeepSeek Harness's default "phone entry"

dsh-im is a plugin that connects WeChat, Feishu, DingTalk, QQ, WeCom, Slack, Telegram, Discord, and WhatsApp to DeepSeek Harness (DSH) on your computer, so a message from your phone can direct the agent to get work done. It hit 3,266 stars within 84 days. Rivals that started the same day had 59. The idea is one anyone could have had. It won because its author had already built the same thing before everyone else.

4:40 a.m. (Beijing time)

DeepSeek open-sourced DSH in Aug 2026 13, an agent framework where "everything is a plugin." It hit the Hacker News front page the same day (747 points) and now has about 270k stars.

At 15:4 a.m. Beijing time on Aug 40, GitHub user xmanrui created a new repo. The first commit was "Add unified IM channel plugin," a unified IM channel plugin. 8 minutes later came a bilingual Chinese/English README and screenshots. Within the next day the commit log reads: "Add DingTalk," "Add QQ bot channel," "Add WeCom QR bot channel." DingTalk, QQ, and WeCom were connected one after another.

Aug 16, 10:30 AM, the Bilibili account "OpenClaw中文社区" posted a video titled "Released a plugin that connects WeChat, Feishu, DingTalk, QQ, and WeCom to the deepseek harness." As of today it has 20K views and 376 favorites. The next day it posted another, titled "Spent 1B tokens building a... plugin." The account's bio reads "OpenClaw-bot-review author," which means xmanrui himself.

Someone else built nearly the same thing that same week. On Aug 15, the DeepSeek tech community on CSDN published "Turning WeChat and Feishu into AI agent entry points." Its opening states the need well: "We usually use AI coding assistants sitting at the computer... but many scenarios happen on the phone... dsh is missing a 'phone entry.'" But that article covers another person's plugin, dsh-im-bot, which only connects WeChat and Feishu and requires running an install script. That repo has 2 stars today.

What it is

In one line: turn your chat apps into a remote control for the agent on your computer. You're out, you message the bot in WeChat "help me look at this error," and the DSH on your home computer reads files and runs commands, then sends the result back to the chat.

Install is through DSH's built-in plugin command, dsh plugin --profile web add -w @xmanrui/dsh-im. After a restart, an "IM Bot" page appears in settings. Most channels set up by QR code: Feishu, DingTalk, WeCom, and QQ auto-create a bot when you scan with your phone; WeChat binds by QR code; WhatsApp links a device by QR code. Slack, Telegram, and Discord need a token pasted in. iMessage and experimental Matrix support have been added as well.

The README's channel table shows the author actually used it. WeChat replies are split into segments of 1,800 characters and show "typing..." while the answer is pending. DingTalk streams output through an AI card. On Discord, the first @mention opens a thread, and after that no more @s are needed. WhatsApp responds only to chats with yourself by default, so friends' messages don't trigger the agent. Each channel can host multiple bots, each bound to a different workspace and model. Images can come in, and result files can go back out. None of this is something you'd think of on day one; it's what users asked for.

Over three months the version went from 0.16 to v5.10.0, and the most recent commit was 11-05.

Why this one

The author didn't start from zero. Before DSH, the hottest open-source agent was OpenClaw, and OpenClaw's whole approach was plugging agents into chat apps. xmanrui's earlier project, OpenClaw-bot-review (2,025 stars), is a panel for viewing OpenClaw bots, models and session state, and his Bilibili account is literally called "OpenClaw中文社区". DSH itself is built on Cordis, the same foundation as Koishi, the long-running Chinese chatbot framework. The other foundation is WeChat: in 2026 March, WeChat launched a ClawBot plugin entry and opened the iLink Bot protocol, so connecting a personal WeChat to an agent went through an official channel for the first time, with no simulated login and no ban risk (SegmentFault's 8-26 integration article). So when DSH launched, his main job was moving things he already knew over to it, which is probably why he could connect five channels in one day (inference).

One plugin for everything. zhuiyueya/dsh-im-gateway, launched the same day, claims to connect 20 platforms and has 59 stars today; dsh-lark, Feishu only, has 60 stars. Readers want "install one and you're done", and Chinese users are often on WeChat, Feishu and DingTalk at the same time. By Aug 26, SegmentFault's "How DeepSeek Harness Connects to WeChat: Four Routes" already listed it as a "mainstream integration option" (about 393 stars then), concluding that "ordinary users should pick dsh-im first".

The author promoted it himself, then the ecosystem took over. 8-16 and 8-17 were two Bilibili videos from him; 8-19 was posted in DSH's official discussion forum under "Show Your Plugins", and 8-29 was another post. 8-22 On X, a Chinese resource account reposted it: "Basically installs your own AI agent into WeChat and Telegram." 8-27 The Bilibili creator "全栈二把刀" made an episode, "DSH插件之IM接入", 8-21 and someone on Zhihu wrote a hands-on test of nine channels. After that, directories like awesome-dsh-plugin (22.6k stars) and dsh-plugin.org listed it. At that point it had become the default answer to "connect DSH to WeChat", and anyone searching later finds this one.

Growth has been steady. 10-17 was at 2,054 stars, 11-06 at 3,266 stars, gaining forty-some to over a hundred a day, and it has never made GitHub Trending. Over the last 7 days: +486; the first 7 days: +456. The biggest single day was 11-01 at +150, the same morning a 2.11M-view Bilibili full DSH tutorial went up, and several DSH plugins rose together, possibly pulled along by that video (inference; I haven't verified the video mentions it).

333 forks, which is 10% of stars, high for a plugin. It suggests many people are taking it, modifying it and deploying it themselves, not just bookmarking it.

Who it's competing with

Almost all similar DSH plugins handle only one channel, an order of magnitude behind it. The most telling entry is dsh-wechat: it arrived nearly two months after the platform launched, handles just one channel, and still reached 400 stars, now gaining 49 a week. So the need to "connect to chat apps" hasn't been fully covered in DSH, and deep single-channel plugins still find users.

If it were you

Verdict: don't build yet another DSH multi-channel IM plugin. This slot is already an order of magnitude ahead and still updated weekly. The lesson is the slot itself: every hot agent platform that launches creates demand to connect it to the chat apps users already have, one of the hardest-demand slots around, and whoever has done it once can do it again much faster.

Good fit for: developers who have already added IM to OpenClaw, DSH or another agent and have channel adapter code ready; people familiar with the WeChat, Feishu, DingTalk and WeCom open platforms.

Bad fit for: people who have never built an IM bot and would have to learn each platform's docs from scratch (by the time you finish three channels, the default answer is already set); anyone hoping to earn stars with a "toy" single-channel version.

How to do it:

1. First, make the channel adapter layer framework-independent (receive messages, streaming replies, message splitting, sending images and files, permission allowlists). That's the part you can carry to the next platform.
2. Watch for the next "everything is a plugin" agent platform. Be live from launch day to the next day; the first version should at least cover WeChat and Feishu, work with a QR code scan, then add a channel a day.
3. Go deep where DSH's version is thin. dsh-wechat has shown that deep single-channel plugins have an audience. Look at the 77 open issues in dsh-im and pick one channel or scenario (group collaboration, WeCom self-built apps, or overseas channels like LINE and KakaoTalk that it doesn't cover) and make it the best option.
4. Put the full package name and a one-line install command on the first screen of the README, and a channel table that explains how each platform is connected and how replies work.

How to get users: Same as it did. On launch day, record a Bilibili demo titled something like "Connect WeChat and Feishu to XX". Post in the plugin showcase section of the platform's official forum. Submit to awesome lists and plugin directories. Then wait for the first wave of Chinese reviews (CSDN, SegmentFault, Zhihu) to call you a "mainstream option".

When to move: within a week of a new platform's launch. dsh-im was called a "mainstream integration option" within 11 days of launch, and everyone who comes later is measured against its default position.

Signs you've misread it: ① DSH's official team ships its own IM connector, and this slot shrinks as a whole. Before you build "the platform's missing piece", check whether the platform will fill it itself. ② Single-channel plugins like dsh-wechat stall at a few hundred stars, which means DSH has run out of gaps. ③ The next hot agent platform ships with a built-in chat app entry, and this slot no longer exists.

Data and sources: GitHub repo page, commit history and daily star counts (10-17 to 11-06); DSH official discussions #3435 (8-19) and #5080 (8-29); Bilibili "OpenClaw中文社区" 8-16 and 8-17 videos, "全栈二把刀" 8-27 video; CSDN DeepSeek Technical Community 8-15 article; SegmentFault 8-26 article; X @grgerwcwetwet 8-22; Hacker News (DSH launch thread).