Everyone Ignores This Hack. It Might Be the Best One Ever Pulled Off.
In this article
> **Bottom line:** The best-kept secret in software isn't a clever prompt or a new framework — it's that GitHub Actions' free tier (2,000 minutes/month on free accounts, unlimited on public repos) can run as a complete backend: cron scheduler, scraper, database, and API, for $0.
The technique, popularized by developer Simon Willison under the name "git scraping," stores data as commits in a git repo instead of rows in a database, then serves it straight from `raw.githubusercontent.com`.
Projects tracking everything from wildfire perimeters to congressional stock trades run entirely on this stack.
If you're still spinning up a Postgres instance and a cron server for a side project that checks a website twice a day, you're paying for infrastructure that already exists inside a tool you use every day for free.
I've paid for a $12/month DigitalOcean droplet for three years to run a scraper that checks one webpage twice a day and sends me a text.
I felt smart about it. Twelve dollars a month, that's nothing, right?
Then I watched a YouTube walkthrough of someone doing the exact same job — scheduled fetch, transform, alert — using nothing but a GitHub Actions workflow file and git commits.
Zero dollars. Zero servers. Zero maintenance. I'd been renting a house when I could've been squatting in a mansion nobody was using.
That's the hack. And it's been sitting in plain sight since GitHub Actions launched in 2019.
The Setup: Git Was Never Just for Code
Here's the part almost every developer misses: **a git repository is an append-only log with perfect version history, free hosting, and a built-in diff engine.** That's also, functionally, a description of a time-series database.
Simon Willison — creator of Datasette, former co-founder of Lanyrd, a name most backend engineers will recognize — named this pattern "git scraping" back in 2020. The idea is almost insultingly simple.
You write a GitHub Actions workflow that runs on a cron schedule, fetches some data (a webpage, an API, a PDF), saves it to a file, and commits it.
Every commit is a snapshot. The commit history *is* your database.
People have used this to track California wildfire perimeters, UK government press releases, NYC restaurant inspection data, and stock trades disclosed by U.S. senators.
None of these projects run a server.
None of them pay for a database. They run on GitHub's compute, store data in GitHub's storage, and serve it through GitHub's CDN via raw file URLs or GitHub Pages.
Total infrastructure cost: **$0**, up to the free-tier limits.
Everyone Treats GitHub Actions Like a CI Tool. That's the Mistake.
Ask ten developers what GitHub Actions is for and nine will say "running tests" or "deploying on merge." That's not wrong, it's just incomplete — and the incompleteness is costing people money and time they don't need to spend.
The mainstream take is that CI/CD tools are for CI/CD. Fair enough on paper.
But that framing blinds people to what Actions actually *is* under the hood: a free, scheduled, containerized compute environment that can run any script, on any interval, triggered by any event, with authenticated write access back into your repo.
That's not a CI tool. That's a cron server, a job queue, and a data pipeline wearing a CI tool's name tag.
The reason this gets ignored isn't that it's hard. It's that it doesn't *look* like infrastructure.
There's no dashboard with a database icon on it, no "Deploy to Production" button, no bill that reminds you infrastructure exists.
It's a YAML file sitting quietly in `.github/workflows/`, and most developers walk past it the same way they walk past a fire extinguisher — assuming it's for a different emergency than the one they're having.
**The best hacks are never hidden. They're just mislabeled.**
The Framework: The Four-Layer Free Stack
Once you see GitHub Actions as infrastructure instead of tooling, the whole shape of a backend collapses into four layers you already have access to.
I call it the **Free Stack**, and it maps almost exactly onto what you'd otherwise pay for.
Layer 1: Trigger (replaces your cron server)
GitHub Actions supports `schedule` events using standard cron syntax, plus `repository_dispatch` for webhook-style triggers.
You get scheduled execution without provisioning anything — no EC2 instance babysitting a crontab, no forgotten `cron.daily` script nobody remembers writing.
Layer 2: Compute (replaces your worker process)
Every workflow run spins up a fresh Ubuntu, Windows, or macOS container with Python, Node, and most common tooling preinstalled.
You get ephemeral, isolated compute for free, up to 2,000 minutes a month on free personal accounts, and effectively unlimited on public repositories.
Layer 3: Storage (replaces your database)
This is the part people miss. Committing a JSON, CSV, or SQLite file back to the repo turns your commit history into a queryable, versioned, append-only dataset.
Want to know what changed between Tuesday and Thursday? `git diff` already does that — it's the query you'd otherwise write in SQL.
Layer 4: Serving (replaces your API and CDN)
`raw.githubusercontent.com` serves any file in your repo over HTTPS, cached and fast.
Pair it with GitHub Pages and you've got a static API and a front end, both free, both with no servers to patch at 2 a.m.
Stack those four layers and you've replaced a cron server, a worker, a Postgres instance, an API gateway, and a CDN — the exact bill most solo developers and small teams are quietly paying every month for workloads that don't need any of that muscle.
What This Actually Changes for You
If you're a solo developer or indie hacker, this is the difference between a side project that dies because the hosting bill outlasts your motivation, and one that just keeps running for years because it costs nothing to leave alone.
Willison's own git-scraped datasets are still updating today, some of them years after he stopped actively maintaining the code.
If you're at a company, this is worth a harder look than most teams give it.
I'd bet a meaningful share of the scheduled jobs your team runs — internal reports, data syncs, health checks, alert scripts — are running on infrastructure that costs real money and does a job the Free Stack could absorb entirely.
Not everything.
**Anything handling sensitive data, requiring low latency, or needing real concurrency still needs real infrastructure.** But the boring, periodic, low-stakes automation that makes up a surprising share of internal tooling?
That's exactly what this pattern eats for breakfast.
If you're managing engineers, the interesting signal isn't the money saved, it's what noticing this pattern says about someone.
The developers who spot that a CI tool can be a database are the same ones who'll find the unused capability in whatever new AI agent framework your company adopts next.
That instinct — **squint at the tool you already have before reaching for a new one** — is worth more than most frameworks.
And yes, in 2026 this extends past scraping.
People are scheduling small AI agent jobs — a Claude or GPT-5 call that checks a data feed, summarizes it, and commits the summary — on the exact same free scaffolding.
The pattern doesn't care what's inside the box on Layer 2.
The Bigger Picture
There's a reason this hack stays obscure even though it's been public for six years: it doesn't sell anything.
There's no company behind "use git as your database," no vendor with a marketing budget pushing it into your feed.
It survives entirely on word of mouth and the occasional YouTube deep-dive, which is exactly why it fits "trending on YouTube" better than most funded product launches ever will.
That's the real lesson buried under the technical trick. The most valuable moves in tech are rarely the ones with the biggest launch event.
They're the quiet reuses — someone looking at a tool everyone already has and asking "what else is this secretly capable of?" instead of reaching for their credit card.
I've spent fifteen years in this industry watching people, myself included, default to buying a new tool before fully using the one already open in another tab.
The Free Stack isn't going to replace real infrastructure for anything that matters at scale.
But it's a pretty good reminder that most of us are under-using what we already pay for, or in this case, what we're not even being asked to pay for at all.
What's the tool sitting in your stack right now that you're only using for 10% of what it can actually do?
---
