What free-for-dev's Deletions Teach About Free Tiers

The most useful thing about free-for-dev isn't the list. It's the three rules at the top that decide what gets on it. A service must be a hosted offering, not self-hosted software. It must have a genuine free tier, not a trial. And if that tier is time-bucketed, it has to last at least a year. That filter is why a Markdown file started by R.I. Pienaar in 2015 now sits at roughly 134,000 stars with contributions from more than 1,600 people, and why it still trends on GitHub a decade later: it's the closest thing developers have to a ledger of who actually gives away compute, and who only pretends to.

A free tier is a liability, and vendors know it

Read the list as a time series rather than a catalogue and a pattern jumps out. The entries that got removed tell a better story than the ones that remain.

Heroku's free dynos and free Postgres went away on November 28, 2022, and for a generation of developers that was the end of an era: the git push heroku main side project was the default onboarding flow for a decade. Railway dropped its free plan in 2023. PlanetScale retired its Hobby tier in April 2024, which stung more because people had put real production data on it. Fly.io stopped giving free allowances to organizations created after October 7, 2024, keeping them only for grandfathered accounts.

None of those companies were being stingy for sport. A free tier is a marketing cost that scales linearly with abuse: crypto miners, spam relays, and abandoned databases that nobody pays to delete. When the cheap-money era ended, the tiers that had been justified as top-of-funnel spend got re-evaluated as infrastructure liabilities. free-for-dev's pull request history is where those decisions show up as one-line diffs.

That's the first editorial takeaway: treat any free tier as a revocable grant, not a foundation. Build on it, but keep your exit cheap.

Who still gives it away, and why

The survivors fall into three buckets, and the bucket tells you how safe the tier is.

Hyperscaler "always free" allocations. Oracle Cloud is the outlier here, offering Arm-based VMs and block storage with no expiry, which is why it's a perennial favorite for self-hosted side projects despite the account-provisioning headaches. AWS, Google Cloud, and Azure all maintain a permanent free slice alongside their 12-month trial credits. These are the safest tiers on the list, because the cost of your usage is a rounding error on their balance sheets and the strategic value of you learning their console is real.

Edge and serverless platforms where idle is genuinely free. Cloudflare Workers gives you 100,000 requests per day on the free plan; Vercel and Netlify keep hobby tiers alive for static and serverless deploys. The economics work because a request-metered product has near-zero marginal cost for an app nobody visits. Contrast that with Heroku's model, where a free dyno was a running process consuming memory whether or not anyone cared.

Developer-tooling vendors whose free tier is the product. GitHub Actions minutes, Sentry's error quota, Grafana Cloud's hosted metrics. For these companies the free tier isn't marketing spend, it's the adoption channel that makes the paid tier inevitable once a team grows. They'll tighten limits, but they rarely pull the tier outright.

If you want a rule of thumb: the more a free tier costs the vendor when you're idle, the more likely it is to disappear. Always-on containers and provisioned databases are the endangered species. Request-metered and quota-metered services are durable.

How to actually use the list

free-for-dev organizes itself into 40-plus categories, from major cloud providers through CI/CD, managed databases, monitoring, email, and DNS. Don't browse it. Search it with a stack in mind and assemble a zero-cost architecture deliberately:

• Compute: a Cloudflare Worker or Vercel function for the request path; an Oracle Always Free VM if you need a long-running process.
• Data: pick a serverless Postgres that bills on compute-hours rather than provisioned instances. The list has several; the ones that survived 2024 all scale to zero.
• CI: GitHub Actions covers most of it at 2,000 minutes a month on private repos, unlimited on public ones.
• Observability: a hosted Grafana or Sentry free quota, wired up on day one so you're not flying blind when the thing unexpectedly gets traffic.

Then do the one thing most people skip: write down the migration path for each piece before you depend on it. If your database vendor sends a ninety-day deprecation email, do you have a pg_dump target? If your edge platform changes its request cap, is your code portable or is it stapled to proprietary bindings? The list tells you what's free today. It can't tell you what's free in eighteen months, and its own history suggests the answer is "less."

The uncomfortable bit about curation

There's a legitimate critique of any awesome-list: the incentive to get listed is a marketing incentive, and 1,600 contributors includes a lot of developer-relations staff. free-for-dev mitigates this better than most because its rules are mechanical and the README is honest that inclusion is opinionated toward infrastructure developers. But a listing is not an endorsement, and the list doesn't track reliability, support quality, or how aggressively a vendor nudges you to a credit card. Cross-check against the project's issue tracker and recent PRs, where "removed X, no longer free" commits are the most valuable signal in the repo.

Verdict

free-for-dev is production-grade as a reference and a trap if read as a promise. The durable free tiers in 2026 are the ones where your idle usage costs the vendor nothing: edge functions, quota-metered tooling, and hyperscaler always-free slices. Everything else is borrowed time. Use the list to prototype for free, keep every component replaceable, and read the deletions as carefully as the additions.