I built this Bitcoin dashboard for a slightly nerdy reason. I wanted to understand GitHub's built-in AI model library, and reading the docs is not the same as shipping something with it. So I shipped something. The thing I learned along the way is worth more than the project itself, and it has very little to do with Bitcoin.
Here is the short version of what I wanted to test. GitHub quietly ships a service called GitHub Models: a catalogue of frontier models from OpenAI, Meta, DeepSeek, Mistral and others that you can call straight from an automated workflow with nothing but your own repository's token. No API key to provision. No billing to set up. No separate account. It runs on a free tier that is rate-limited, and that limit is generous enough that anything short of a production workload never touches it. I had read about it. I wanted to feel it work.
So I picked a subject I find genuinely interesting, cryptocurrency, and a format I think is badly underused, the live dashboard, and I wired them together. What came out the other side changed how I think about content. Let me make the argument first, then show you the machine.
The reason a dashboard beats a blog post is not the one you would guess
If you run marketing, or you run the business that pays for it, you have had this conversation. We should publish more. We need to show up in search. The blog has gone quiet. All reasonable. All slightly beside the point now.
A blog post is read once and then it slides down the feed and dies. You write it, it has a good week, and it spends the rest of its life as an entry in a sitemap. A dashboard does not behave like that. A dashboard is a destination. It updates itself, so it stays relevant without anyone on your team touching it. It earns links and embeds because it is useful rather than promotional. And it proves you understand your market in a way no listicle can, because you cannot fake a working model of the thing your customers actually care about.
A live, public dashboard is one of the most underrated pieces of content a business can make. Almost nobody is making them. The reason used to be cost. That reason is gone.
That last point is the one that matters, and it is the one I went and proved. The marginal cost of building and running the thing you are reading about is, to the nearest round number, zero. Free hosting. Free data. Free inference. The only scarce input was an afternoon. When the cost of a category of content falls through the floor and almost nobody has noticed, that is not a gap in the market. That is the market.
Whatever you sell, sim racing gear, coffee, conveyancing, legal cover, there is a live number your audience checks, argues about, or worries over. A dashboard is the content type that owns that number. So here is how I built one for under the price of a coffee, and how you would build yours.
No backend, and that is the entire point
There is no server. There is no database. There is not even a build step. The whole site is plain HTML, SVG and JavaScript that loads in your browser and then fetches its data from free public APIs and renders it on the spot. Open the dashboard and your own machine pulls the price, the sentiment, the flows and the network stats. Nothing of mine is sitting in the middle.
This is not a shortcut, it is the design. No backend means nothing to break at 3am, nothing to pay for, nothing to patch. It deploys to GitHub Pages straight from the repository. Every panel is written to fall back to "n/a" if a feed is slow or blocked, so one flaky API never takes the page down. It just shows a small dash where a number would be and carries on.
The spiral: a full market cycle you can read in five seconds
The picture at the top is generated by a single Python script with no dependencies. It plots Bitcoin's price against the Power Law model on a polar map where one full turn is a four-year cycle. Angle tells you where in the cycle you are. Radius is price on a log scale. The orange band is the channel Bitcoin has tracked for fourteen years: floor, fair value, and the cycle-top region. It is built to be read in five seconds and then stared at for five minutes, which is the only honest test of a chart.
The dashboard: from "number go up" to an actual read
The dashboard is the live view. It pulls a dozen signals and renders stat tiles, a stack of charts, and a verdict banner that blends six indicators into one plain-English call: accumulate, lean buy, neutral, lean sell, distribute. The aim is not to look clever. The aim is to take someone from "is the number going up" to "here is where we sit in the cycle, and here is what that has tended to mean before."
The part I am proud of: an AI note that does not lie
Every six hours an automated job wakes up, and this is where GitHub Models earns its place. The pipeline does five things, in order:
- Refresh the data. Pull real monthly closes, scrape spot-ETF flows, fetch the institutional headlines.
- Redraw the spiral from the fresh numbers.
- Compute the signals. Four independent reads, all in plain code: valuation (where price sits in the channel, and how that ranks against every month since 2011), trajectory (30 and 90-day change, plus the Mayer Multiple), sentiment (Fear and Greed, now versus a month ago) and demand (7 and 30-day ETF net flow).
- Write the note. GitHub Models turns those numbers into three or four calm sentences.
- Commit and deploy. The refreshed data and image are saved back and the site redeploys. It is never more than six hours stale, and no human touched it.
The discipline in those last two steps is the whole trick, and it is the thing most people get wrong the moment they bolt an AI onto their data. So let me be blunt about it. The model is a writer, not a calculator. Every figure it sees is worked out in code first, and the instruction it is given forbids it from inventing or predicting a single number. So the words always agree with the dashboard. It never hallucinates a price target. And if the live feeds are down, it falls back to a note built from the last good numbers rather than making something up.
An AI model is brilliant at turning a pile of numbers into a sentence a human wants to read. It is terrible at being the source of those numbers. Keep those two jobs apart and the result is reliable. Blur them and you have shipped a confident liar with your logo on it. That distinction is the difference between AI that earns trust and AI that quietly erodes it, and it is worth more than any model upgrade.
What it costs you: nothing
Hosting is free. The price, sentiment and network data are free. The AI inference is free, on a tier that four calls a day does not come close to troubling. There is no line item. The barrier to this kind of content was never the technology, and as of about eighteen months ago it is not the cost either. The only thing standing between you and a dashboard your customers bookmark is the decision to build one.
Take it, and point it at your own numbers
This is open source under the MIT licence. Fork it, take Bitcoin out, point the same pipeline at whatever number your audience actually cares about, and you have a self-updating, AI-narrated dashboard of your own. That was the experiment. As far as I am concerned, it worked, and the conclusion is sitting in your hands.
If you want the engine room, the technical reference has the full maths: the Power-Law constants, the spiral mapping, every indicator formula, the exact verdict weighting, the data sources, and how it all runs without a single API key.
Educational only. Not financial advice.