ChatGPT-On-CS drives marketplace chat from an Electron shell, across many platforms
AI customer-service bot for Pinduoduo, Qianniu and Douyin Store: auto-replies to customer inquiries, product questions and after-sales disputes, with unified support across WeChat, Xiaohongshu, JD, Douyin, Bilibili and Weibo. Pluggable with LLMs like DeepSeek and Qwen, and supports a custom knowledge base.
At a glance
- What is it?
- ChatGPT-On-CS is a TypeScript Electron desktop client that answers buyer messages on Taobao, Pinduoduo, Douyin, Xiaohongshu and WeChat through an LLM. The interesting part is the platform coverage, not the model choice.
- Who is it for?
- Operators running a handful of storefronts who want model-written replies without buying a CRM should take it, provided they can live with the terms the project publishes. Anyone who needs seat permissions, shift handover, an audit trail or a documented install path should skip it and start from the vendor site instead.
- Can I use it commercially?
- Yes, with strict conditions. AGPL-3.0 is a network copyleft licence: if people use a modified version over a network, for example as a hosted service, you must offer them its source code under the same licence.
- Is it still maintained?
- Yes. The repository last received commits 36 days ago.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- cs-lazy-tools/ChatGPT-On-CS4,610 ★ · +246
| Repository | 2026-09-10 | 2026-11-15 |
|---|---|---|
| cs-lazy-tools/ChatGPT-On-CS | 4.4k | 4.6k |
246 Stars in the last 66 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
CS means customer service, not Counter-Strike
The name invites a wrong reading, so settle it first: this is a marketplace buyer chat client. package.json describes it as "多平台智能客服,允许使用 ChatGPT 作为客服机器人", and the product it wraps is an Electron desktop app with a React renderer built through the electron-react-boilerplate layout (.erb/, electron-builder.yml, .vscode/). Nothing in the repository touches a game, a game process or game memory.
The problem it addresses is narrow and expensive. A single seller on Pinduoduo, Taobao, Douyin or Xiaohongshu gets buyer questions in the platform's own chat widget, and human agents answer them one at a time during working hours. The buyer waits, and off-hours messages sit unread. The project exists to keep a language model answering that queue while a human can take over any single conversation. That audience is a shop owner or a small operations team, not a platform engineer.
Platform adapters are the real architecture
The shipped list covers WeChat, Qianniu, Bilibili, Douyin enterprise accounts, Douyin, Doudian, Pinduoduo, Weibo chat, Xiaohongshu professional accounts, Xiaohongshu, Zhihu, and the Jingmai and Feige widgets. The roadmap's completed items include browser multi-instance support, Chinese path handling, and manual takeover detection.
The completed list includes 浏览器多开支持, which means several logged-in storefront sessions run side by side inside one desktop app.
The second layer is behaviour policy and scenario agents, described in the demo captions as 场景 Agent 管专项,行为策略管全店. One agent handles a narrow case, the policy governs the whole store. A third layer, the 场景训练台, fakes an inbound conversation so a reply can be reviewed with its source attached, without waiting for a real customer.
No install command exists, so start from the vendor page
The README contains no clone line, no package install and no run command. It points readers to the hosted product at jinxiaoai.com and to an app catalogue at jinxiaoai.com/apps for per-application documentation. The repository's GitHub releases stop at v1.4.5 from 2024-09-07.
For a developer working from source, the package.json scripts are the whole story. Build both the main process and the renderer:
pnpm run buildThat expands to build:main and build:renderer, each a webpack production run under cross-env NODE_ENV=production. Start the app with:
pnpm run startThe start script first runs .erb/scripts/check-port-in-use.js, so a stale dev server is caught before Electron launches. Note the postinstall chain, which is where a first build can bite:
ts-node .erb/scripts/check-native-dep.js && electron-builder install-app-deps && pnpm run build:dllNative dependencies are checked and rebuilt, then a development DLL bundle is produced. Packaging into a distributable runs clean.js, the build, and electron-builder build --publish never, meaning the package step does not publish.
A reserved skill page and a roadmap that never shipped
Two details in the README matter for a buyer deciding whether to commit.
First, the skill market link is commented out with a note stating that as of 2026-08-27 the URL jinxiaoai.com/skills still returns 404, and the comment warns that adding the link to the top navigation before the page exists costs more than omitting it. A capability page is reserved but not live.
Second, the roadmap marks three features as in development: automatic replies on Douyin live streams, automatic content publishing across platforms, and local large model support. The last one matters most for anyone planning to run buyer conversations without sending order data to a hosted API. It is not available, so every query goes to GPT, Tongyi, Wenxin or DeepSeek.
A third detail catches only a careful reader: package.json points its bugs and repository fields at lrhh123/ChatGPT-On-CS, not the cs-lazy-tools organisation the repository now lives under. Issue links built from that metadata land somewhere other than the canonical project.
White-label resale is solicited while the code ships AGPL-3.0
The repository is licensed AGPL-3.0, and the README restates the terms in four points: personal use is free with no restriction, commercial use requires contacting the project for a commercial licence, code modifications must be open sourced with copyright notices retained unless a commercial licence is granted, and the full text governs. For an operator running their own shop on their own machine, that last sentence is the one that decides the question.
Commercial use requiring a separate licence is a reason to read the licence file before planning a hosted fork for a client, because AGPL reaches network interaction. This is not legal advice, only the reading of two documents sitting in the same repository. Note also that the project was last pushed on 2026-10-10.
Session loss and unreviewed buyer answers
The repository documents no recovery procedure, no retry policy and no alerting for a platform connection that stops answering.
Automatic replies on a live marketplace carry a second risk that no code setting removes. A wrong answer from the model about a return policy or a shipping window goes to a real buyer. The completed list shows 延时随机回复功能, a delay plus randomisation meant to look less robotic, which also means a bad reply is delivered with a delay nobody can retract. The 人工接管自动检测 item detects when a human takes over, but detection of takeover is not detection of a wrong answer.
Knowledge base files are the mitigation: upload your own documents and the answers stop being improvised. The demo caption makes the claim that the knowledge base is learned from real conversations rather than written by hand. That still leaves the accuracy of any single reply unverified.
Inside the channel rather than beside it
The comparable category is a hosted customer service platform such as Zendesk or Intercom with an AI agent bolted on. The approach differs in one decisive way. Those products sit beside the channel: agents work in a ticketing system and messages are forwarded or integrated through an official API. ChatGPT-On-CS connects directly to the marketplace chat on platforms such as Pinduoduo and Xiaohongshu.
The cost sits in the same place. A hosted helpdesk has permissions, transcripts, escalation rules and audit history. This has an Electron window, a knowledge base and a data centre screen counting how many conversations the model handled and how many were handed to a human. For a single seller whose buyers ask the same twenty questions, the swap is fair. For a team with seat management, shift handover and compliance requirements, it is not.
Editorial conclusion
Operators running a handful of storefronts who want model-written replies without buying a CRM should take it, provided they can live with the terms the project publishes. Anyone who needs seat permissions, shift handover, an audit trail or a documented install path should skip it and start from the vendor site instead. Verify first whether the specific marketplace you depend on is on the supported list. Check the pnpm start path before committing.
Frequently asked questions
Is ChatGPT-On-CS a Counter-Strike bot or a game cheat?
No. Despite the name, ChatGPT-On-CS is an e-commerce customer service client for marketplace chat, described in package.json as a multi-platform customer service robot using ChatGPT. It has no connection to Counter-Strike, game memory or live match automation.
How do I install ChatGPT-On-CS?
The README gives no install command and no download link for a self-hosted binary. It directs readers to the hosted product at jinxiaoai.com and per-application documentation at jinxiaoai.com/apps. From source, the documented entry points are the pnpm run build and pnpm start scripts in package.json.
Which marketplaces does ChatGPT-On-CS support?
The shipped list covers WeChat, Qianniu, Bilibili, Douyin enterprise accounts, Douyin, Doudian, Pinduoduo, Weibo chat, Xiaohongshu professional accounts, Xiaohongshu and Zhihu, plus the Jingmai and Feige widgets.
Can ChatGPT-On-CS run on a local large model?
Not yet. Local large model support is listed as in development on the project roadmap. Until it ships, answers go to hosted models such as GPT, Tongyi, Wenxin or DeepSeek, so buyer and order data leaves the machine.
What license is ChatGPT-On-CS released under?
The LICENSE file is AGPL-3.0. The README states personal use is free, commercial use needs a commercial licence from the project, and modifications must stay open sourced with copyright notices unless such a licence is granted.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/cs-lazy-tools-chatgpt-on-cs)Star history chart for your README
[](https://hysenlabs.com/projects/cs-lazy-tools-chatgpt-on-cs)Was this useful?