AIProductHome: a hand-curated catalogue of AI tools, and what it takes to keep one alive

There is no software in this repository, and that shapes everything

The repository description is the shortest honest summary available: collect various AI products, in Chinese and English, with the Chinese half reading as a catalogue of AI tools and products. The tree has four Markdown files, an `img/` directory and a `.github/` directory. There is no source tree, no package manifest, no build configuration and no test suite.

So calling it a framework or a platform would be wrong. It is a documentation repository whose primary artefact is a README that functions as a directory. The software it describes lives in other places, which is why every row in every table is a link out rather than an installable thing.

That has a practical consequence for anyone evaluating it. You cannot depend on it the way you depend on a library, and there is no version to pin. What you can take from it is the curation method. The icon legend, the section structure and the intake rules are transferable to any directory project, whether that directory covers AI tools or something else entirely.

Three mirrors, because one URL is not reachable everywhere

The changelog entry dated 2026-06-22 lists three destinations for the same content, and the reason it lists three rather than one is the clearest signal about the audience:

2026-06-22
- 在线网站：https://eternaldeath.github.io/AIProductHome/
- 国外仓库：https://github.com/Eternaldeath/AIProductHome
- 国内仓库：https://gitcode.com/qq_34902437/AIProductHome/overview

There is a hosted site built from the repository, the GitHub repository itself, and a GitCode mirror. GitHub Pages is slow or unreachable from inside mainland China, which is the reason a third host exists rather than an aesthetic choice. Any directory project serving a Chinese-speaking audience runs into this, and most of them do not solve it.

The repository also carries a WeChat group badge, an RSS badge wired to the repository's commits atom feed, and star and fork badges. The WeChat group is described as the place to join by adding a contact and noting that you want the AI tools group. The commits feed is the more useful of the two for following along, since it reflects the actual edit history.

The icon legend is the entire editorial signal in the tables

Every table row carries a cell of icons, and a legend at the bottom of the README decodes them. The legend maps a flask icon to a testing-stage product, a plane to needing a proxy to reach, a symbol to free use, a money bag to paid use, both together to free and paid mechanisms coexisting, an open lock to open source, a download arrow to a plugin, and a page icon to being based on a paper.

This is a compact and genuinely useful schema. A free, open source, paper-based tool is a different proposition from a testing-stage product that needs a proxy, and a table without those distinctions forces the reader to open forty links to find out. Encoding the distinction in one glyph is the single best idea in the repository.

It also has a cost. Icons age badly. The Claude entry, for instance, carries a note that the tool is limited to the United States and United Kingdom, and the Google chat entry describes a model generation that has since been superseded by several. Neither can be corrected without someone noticing and filing an issue. Section five of the contribution list is help correct errors, which tells you the maintainers know this and rely on readers to do it.

Section structure: relay services, search engines, models, applications

The categories are ordered from infrastructure upward, which is a deliberate editorial choice rather than an accident of how the file grew.

The first category is relay services, currently a single entry for a service that offers an OpenAI-compatible API endpoint giving unified access to Claude, GPT, Gemini, DeepSeek and Grok, positioned for developers in China who need those models without a VPN. Second is AI search engines, with Bing's integrated chat, Perplexity with its source attribution, an academic search tool that surfaces related papers and summarises them, an academic GPT variant, an AI-driven research site with a built-in chat feature, and a Chinese AI search engine.

Third is foundation models, and this is where the distinction the maintainers flagged matters most. GPT is listed as the model and ChatGPT as the application built on it, in one row. A Baidu model, an Alibaba model, a Hugging Face open alternative to ChatGPT, a Google chat bot, two Chinese academic models in medicine and law, a ByteDance machine learning platform, an Anthropic chat tool, a product described as writing malware, and a Xiaomi model follow. Fourth is applications, which is where the recent changelog additions land: an AI-native browser, a resume generator, an Android and iOS runtime protection analysis tool, an AI research workflow product with a Zotero plugin, an AI-native film creation workstation, and several image and video tools.

The category names are in Chinese throughout, so a non-Chinese reader gets the structure and the links but has to guess at the descriptors. Three sibling files broaden the scope: `AI NEWS.md` for industry news and trends, `Prompt.md` for prompt engineering, and `AI Tutorial.md` for tutorials on using and combining tools.

Intake is being tightened, and dead entries get deleted

The 2026-06-22 changelog entry carries three warnings, and together they are the most substantive editorial content in the repository.

The first says the repository is entering an optimisation phase and that the logic of each section will be reworked, with the example given being the split between foundation models and model-based application products, such as distinguishing a GPT model from a ChatGPT product. The second says that for free submissions there will be a threshold for selection, preferring mature and reliable products, while tools that have already stopped working will be deleted. The third says the site development and optimisation will continue to improve the user experience.

Translated into practice, that is a move from an open directory toward a curated one. Submission is by GitHub issue, and the README lists the five reasons to open one: self-submission, a resource has stopped working, offering a resource, correcting an error, and an ellipsis for the rest. Of those five, two are about maintaining existing entries rather than adding new ones, which is a reasonable ratio for a list whose main failure mode is rot.

The changelog itself lives inside a fenced block in the README, tagged as a shell script, which is a small inconsistency in an otherwise careful document. It also means the repository has no separate changelog file, so the update history is only visible as part of the README and the changes are not individually revertable.

What it costs to maintain, and who is paying for it

Running a directory of this kind is not a one-time publishing job. Every entry is a promise that the link resolves, the product still works, and the description is still true. A product with an AI research workflow and a Zotero plugin today will have changed its pricing and its feature set within a couple of release cycles, and the free-or-paid icon becomes wrong the moment that happens.

The deletion policy in the changelog is an admission of this. The maintainer explicitly plans to remove tools that have stopped working, which means someone is checking. That is unpaid work, and the repository carries no licence file, which is worth flagging: with no declared licence, the default under copyright law is that others may read the content but not reuse it. For a list meant to be copied from, that is a genuine limitation on how far it can spread.

The repository is not archived and the last push was on 2026-06-22. Open issues number in the low twenties, and the contribution route through issues rather than a CONTRIBUTING file, which is consistent with a project whose real workflow is a maintainer reading submissions. There is no test automation to speak of, because there is no code to test. The only automated artefact is the GitHub Pages build, and the Star History chart at the bottom is generated by a third-party service rather than anything in the repository.

Editorial conclusion

AIProductHome is worth an hour of reading if you want to see how a community catalogue is run, and worth nothing as a technical dependency, because it contains none. The three mirrors are the practical thing to copy: the GitHub Pages site, the GitHub repository, and a GitCode mirror for readers in China. Start with the legend rather than the entries, since the icons are the only editorial signal in the tables and they encode whether a tool is in testing, needs a proxy, is free or paid, and is open source. The 2026-06-22 entry announces that the repository is entering an optimisation phase, that categories will be reworked to separate foundation models from applications built on them, and that free submissions will be screened with a threshold rather than taken by default. The last push was on 2026-06-22.

Frequently asked questions

Official sources

1. Eternaldeath/AIProductHome on GitHub
2. Issues
3. Project website
4. README