In May I published Before Google: My First Two Websites, about the portal a group of us built as students in Jay Patel’s basement in Baroda. One line from it stayed with me: every pattern returns, addressed to a different audience. That’s easy to write and harder to test. So I bought the domain back and rebuilt 123ofnet.com.
The tagline, then and now
The oldest capture the Internet Archive holds of the original site is from October 18, 2000. It preserves the tagline I wrote, typo and all:
“123ofnet.com — An Easy Soltuion To The NET !!!”
Here’s the tagline on the homepage today:
“123ofnet.com - An easy solution to AI !!!”
I fixed the spelling and kept the three exclamation marks.
The original About page said the site “started off with a small dream — to set up a portal that would help any one find ‘anything’ on the net.” Apart from the phone bills, our only real expense was the domain registration: Rs 400. That dream still holds up. The thing people need help finding has changed.
Why bring it back
The honest answer is that the domain was a loose end. After the site went quiet between 2003 and 2005, the registration lapsed. Between 2013 and 2016 the Wayback Machine shows the domain being used by a spam shop. Someone else’s junk was sitting on the first thing I ever built.
The better answer is that the problem 123ofnet was built to solve is back, and it’s bigger than it was.
Same problem, new flood
In 2000, most people needed help getting onto the internet at all. They needed an email address, some webspace and a list of sites worth visiting. A portal gave them all of that on one page.
In 2026, people need help getting things done with AI. Thousands of AI tools launch every month, and few of them say plainly what they do. Someone who wants to fix a resume, caption a video or get a small-business website live has no easy way to know where to start.
So 123ofnet is a portal again. Its first line reads “Find the right AI tool for anything,” and the directory is sorted by the job you want done, not by vendor or model:
- Chat & Assistants (200 tools)
- Writing & Email (220)
- Career & Jobs (269)
- Pictures & Design (261)
- Video & Audio (264)
- Meetings & Work (286)
- Websites & Business (294)
- Build Apps (No Code) (213)
- Developer (366)
- Learning & Research (262)
That’s 2,635 tools across ten categories, each with its own page and its closest alternatives. Every category starts with a short plain-English guide on how to choose, and a handful of quick picks. For example: Grammarly if you want your grammar fixed as you type, Descript for editing talking videos and podcasts, NotebookLM for asking questions about your own documents.
It’s the same posture as the 2000 homepage, busy but organized, and a lot narrower. The original tried to be the whole internet, with webmail, chat, forums, eGreetings, classifieds and a directory. The new version does one thing first.
Curation versus the algorithm
The promise on the homepage is short: “We check them and list the ones that work.” I want to be precise about what that means, because “curated” has become a marketing word.
Every tool URL goes through a verifier before it gets in. The verifier fetches the site and confirms that it loads, that it isn’t a parked or for-sale domain, and that what the site says about itself matches the listing. Anything that comes back as a mismatch goes to a person. The quick picks at the top of each category are chosen by hand. Every Monday a scheduled job rechecks the links and opens an issue listing anything that has broken, moved or changed, and a person decides what happens to each one. Paid “Featured” spots are labeled FEATURED, affiliate links are marked “(aff),” and paying never skips the review.
This is where the May essay’s argument gets practical. PageRank beat the hand-built directory by reading links at a scale no editor could match, and for twenty years the algorithm deserved to win.
AI assistants and answer engines work differently. When someone asks ChatGPT, Claude or Perplexity “what’s a good free tool to make captions for my videos,” the model doesn’t send them to ten blue links. It writes an answer, and it prefers sources that are clean, categorized, current and easy to attribute. In practice, that describes a well-kept directory.
That’s why the revived site is built to be read by machines as well as people. The robots.txt doesn’t just allow AI crawlers. It names them, from GPTBot and ClaudeBot to PerplexityBot and Google-Extended, under a comment that reads “we want to be cited.” There’s an llms.txt that summarizes every category and its quick picks, and an llms-full.txt with every tool, link and one-line description. Every page carries structured data. This is the same discipline I measure with AEORank, applied to my own first website. The audience for a directory used to be a person with a dial-up connection. Now it’s also an agent answering that person’s question.
How it’s built, then and now
In 1997 the stack was Microsoft FrontPage 4.0, table layouts, an FTP client and free hosting on HyperMart, paid for with a banner ad on every page.
In 2026, I’ll admit the free hosting part hasn’t changed much.
- Static site: Eleventy, deployed to GitHub Pages. GitHub Actions rebuilds and deploys on every push to main, then pings search engines through IndexNow.
- Data: a single JSON file holds the directory. It generates the ten category pages, all 2,635 tool pages, a 2,654-URL sitemap, both llms.txt files and the in-page search index.
- Maintenance: a Python verifier for new tools and the weekly link check that opens a GitHub issue.
- Front end: no JavaScript framework and no third-party trackers. Every page has to work with JavaScript turned off.
- Look: the 2001 design, rebuilt from the archived pages: the blue top bar, the orange tagline strip and the pastel box headers. The one hard rule in the repo is “it must look like the 2001 site.” Underneath, it’s semantic HTML that works on a phone.
The biggest difference is who I built it with. In 1997 it was seven students sharing borrowed machines and a dial-up line. This time it was me and AI coding agents, working from a set of written rules in the repo: never invent a URL, never edit the archived originals, and label anything AI-generated. The first commit went in on October 12, 2026. Four days later the directory had grown from 45 hand-checked tools to 2,635 verified ones.
There’s also a small museum: the original site, restored page by page from the Internet Archive, empty image boxes and all. A few old features are coming back with AI, too. The “Free AI Tools” page lists an email writer, a resume fixer, eGreetings and “Make A Website For Free Today,” all marked as coming soon. In 2000 we gave away 20 MEGS of webspace. Now it’ll be small tools that each do one job well.
Closing
In May I wrote that the portal model died for humans in 2003 and came back for AI agents in 2024. Rebuilding 123ofnet turned that theory into a test. Can a checked, human-maintained directory still be useful when there’s a chatbot in every browser tab? I think the answer is yes, because the chatbots need something trustworthy to cite.
The first version had a typo in its tagline and was built by a group of teenagers in a basement. This one has the spelling fixed and looks almost exactly the same. That’s on purpose.
If you’re trying to find the right AI tool for something, start at 123ofnet.com. If a tool on the list has stopped working, tell me. Keeping the list honest is the reason the site exists.