Last week, I wrote a blog post about how I gave new life to an old domain by building a new website to live on that domain. With help from ChatGPT, it only took a few hours to build the site. While I’ll be adding new businesses and events to the site over time, that is currently a manual process and the site is mostly static.
This week, I wanted to take things a bit further. I wanted to build a site that was updated daily – but without any input from me.
Whois tells me I first registered cool-stuff.co.uk in September 1997. It was one of the first domains I registered. It has hosted a couple of very embarrassing early sites that I built, and for a while, it served as the email domain for several members of my family. But since they all moved to GMail, it’s been pretty much dormant. What it has never hosted is what I originally registered it for – a directory of cool things on the world wide web. So that’s what I decided to build.
So here’s the plan:
- A very simple website
- Each day it features a cool website – just the name, a link and a simple description
- An archive page showing previously featured sites
- Auto-generated each day with no manual intervention from me
I decided to stick with Jekyll and Minimal Mistakes as I enjoyed using them to build Balham.org. They make it easy to spin up a good-looking website, but they also have ways to add complexity when required. That complexity wasn’t needed here.
The site itself was very simple. It’s basically driven from a YAML file called coolstuff.yml which lists the sites we’ve featured. From that, we build a front page which features a new site every day and an archive page which lists all the previous sites we have featured. Oh, and we also have an RSS feed of the sites we feature. This is all pretty basic stuff.
As you’d expect from one of my projects, the site is hosted on GitHub Pages and is updated automatically using GitHub Actions.
It’s in GitHub Actions where the clever (not really all that clever – just new to me) stuff happens. There’s a workflow called update-coolstuff.yml which runs at 02:00 every morning and adds a new site. And it does that by asking ChatGPT to recommend a site. Here’s the workflow:
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 |
name: Update Cool Stuff on: schedule: - cron: '0 2 * * *' # Runs at 2 AM UTC workflow_dispatch: jobs: update: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 - name: Install Perl dependencies run: | sudo apt-get update && sudo apt-get install -y cpanminus cpanm -n --sudo OpenAPI::Client::OpenAI YAML JSON::MaybeXS - name: Get a cool website from OpenAI env: OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }} run: | perl .github/scripts/fetch_cool_site - name: Commit and push if changed run: | git config user.name "github-actions" git config user.email "github-actions@github.com" git add docs/_data/coolstuff.yml git diff --cached --quiet || git commit -m "Add new cool site" git push |
There’s not much clever going on there. I needed to ensure I had an OpenAI subscription with credit in the account (this is going to cost a tiny amount of money to run – I’m making one request a day!), and I set up the API key as a secret in the repo (with the name “OPENAI_API_KEY).
The magic all happens in the “fetch_cool_site” program. So let’s look at that next:
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 |
#!/usr/bin/env perl use strict; use warnings; use builtin qw[trim]; use OpenAPI::Client::OpenAI; use YAML qw(LoadFile DumpFile); use Time::Piece; use JSON::MaybeXS; my $api_key = $ENV{"OPENAI_API_KEY"} or die "OPENAI_API_KEY is not set\n"; my $client = OpenAPI::Client::OpenAI->new; my $prompt = join " ", "Suggest a really cool, creative, or fun website to feature today on a site called 'Cool Stuff'.", "Just return the name, URL, and a one-paragraph description of why it's cool. Only return one site.", "The URL should just be the URL itself. Do not wrap it in Markdown."; my $res = $client->createChatCompletion({ body => { model => 'gpt-4o', messages => [ { role => 'system', content => 'You are a helpful curator of awesome websites.' }, { role => 'user', content => $prompt }, ], temperature => 1.0, } }); my $text = $res->res->json->{choices}[0]{message}{content}; my @lines = split /\n/, $text; my ($name, $url, @desc) = @lines; $name =~ s/^\*\s*//; my $description = join ' ', @desc; my $new_entry = { date => localtime->ymd, name => trim($name), url => trim($url), description => trim($description), }; my $file = "docs/_data/coolstuff.yml"; my $entries = LoadFile($file); unless (grep { $_->{url} eq $new_entry->{url} } @$entries) { push @$entries, $new_entry; DumpFile($file, $entries); } |
We’re using OpenAPI::Client::OpenAI to talk to the OpenAI API. From my limited knowledge, that seems to be the best option, currently. But I’m happy to be pointed to better suggestions.
Most of the code is copied from the examples in the module’s distribution. And the parsing of the response is probably a bit fragile. I expect I could tweak the prompt a bit to get the data back in a slightly more robust format.
But it works as it is. This morning I woke up and found a new site featured on the front page. So, rather than spend time tweaking exactly how it works, I thought it would be a good idea to get a blog post out there, so other people can see how easy it is to use ChatGPT in this way.
What do you think? Can you see ways that you’d like to include ChatGPT responses in some of your code?
The website is live at cool-stuff.co.uk.
I like how the first cool-site is https://example.org, published on April 1st.