The Free Tier That Stays
Short version: database free tiers keep dying for an architectural reason rather than a greedy one. A conventional managed database holds RAM whether or not anyone queries it, so a few hundred thousand abandoned side projects become a permanent cost measured in server racks. Layerbase free databases scale to zero after 15 idle minutes and wake on the next connection in a few seconds, which makes that idle fleet nearly free for us to carry. (MariaDB, the one free engine that is slow to wake, sleeps after 30 idle minutes instead and takes about 20 seconds to come back.) Free is 2 databases, 5 GB, 8 engines, no card and no expiry.
In April 2024, PlanetScale removed its Hobby tier and gave free users a deadline to pay or export. As of July 2026 it still has no free tier. Supabase free projects pause after one week of inactivity, which has spawned a small ecosystem of keep-alive hacks: GitHub Actions on cron, uptime pingers, scheduled functions whose only job is touching the database so the vendor does not put it to sleep (and if a paused project sits more than a year, the restore path is gone). Redis Cloud's free tier is 30 MB. ClickHouse Cloud never had a perpetual free tier, only a trial. Meilisearch Cloud offers 14 days.
Plan details above are from public pricing pages as of July 2026 (PlanetScale, Supabase, Redis Cloud, ClickHouse Cloud, Meilisearch Cloud).
Side by side, that is the state of the free database tier:
| Vendor | What free gets you today |
|---|---|
| PlanetScale | Nothing. Hobby was removed in April 2024 and has not come back |
| Supabase | Projects pause after one week idle, and the restore path is gone past a year |
| Redis Cloud | 30 MB |
| ClickHouse Cloud | A trial only, never a perpetual free tier |
| Meilisearch Cloud | 14 days |
| Layerbase | 2 databases, 5 GB, 8 engines, sleeps at 15 idle minutes and wakes on connect |
If you have been a developer for more than a few years, you have lived this cycle: adopt a free tier, build the side project, watch the pricing-update email arrive. The pattern repeats often enough that "will the free tier exist in two years" is now a legitimate evaluation criterion for a database vendor.
Free tiers die for a boring reason
It is not greed, and it is mostly not investor pressure either. It is architecture.
A conventional managed database is a process that runs all the time. It holds RAM whether or not anyone queries it. Multiply by a few hundred thousand free databases, most of which are abandoned side projects, and the free tier becomes a permanent tax measured in server racks. Every vendor that kills or shrinks a free tier is making the same discovery: idle databases on always-on infrastructure cost real money, and the marketing value of free does not scale with the fleet.
The defenses vendors reach for are all versions of the same move: make idle databases cheaper by making them worse. Pause them after a week. Cap them at 30 MB so they fit in a shared instance. Replace the tier with a trial so nothing idles longer than 14 days. Each of these is rational. Each also breaks the actual use case of a free tier, which is a database that sits mostly idle and works instantly when you come back to it.
Scale-to-zero changes the economics
Layerbase free databases sleep after 15 minutes of idle. The difference between this and Supabase's weekly pause is what waking looks like: there is no dashboard button and no restore flow. You reconnect, and the database is back in a few seconds. Your client sees a slightly slow first connection, not an error. MariaDB, the one free engine that is slow to wake, sleeps after 30 idle minutes instead and takes about 20 seconds to come back. No keep-alive cron job required, and running one would only defeat the point.
That mechanic is the entire reason this free tier is durable. A sleeping database consumes no RAM and no CPU; it is a few gigabytes of storage, which costs us close to nothing. We do not need your abandoned side project to wake up and convert, and we do not need to pause-and-punish it either. The idle fleet that kills other free tiers is, on this architecture, almost free to carry. We are not promising to be generous against our own interests. The interests line up, which is the only kind of free-tier promise worth trusting.
What free actually includes
- 2 databases and 5 GB of storage. Enough for real projects, not a demo allowance.
- 8 engines: PostgreSQL, MariaDB, SQLite, DuckDB, libSQL, Redis, Valkey, and TypeDB. A relational database and a cache, the standard two-piece stack, fits inside the free tier with room left over.
- 1 branch per database, so you can fork a writable copy before a risky migration.
- TLS and IP allowlists included. Security is not the upgrade lever.
- No card, no expiry. Sleeping after 15 idle minutes and waking on reconnect is the only free-tier-specific behavior you will notice.
And the guarantee underneath it: we never delete your data out from under you. Every destructive path takes a backup first, deletion is refused if a final backup cannot be secured, and the most recent backup is never pruned. That applies to free databases, not just paying ones.
The honest caveats
Free databases sleep, so the first query after 15 idle minutes takes a few extra seconds; paid plans can pin databases always-on. Some engines (ClickHouse, QuestDB, Qdrant, Weaviate, InfluxDB) need more resources than a free slot provides and live on the Pro plan. And if your project takes off, we obviously hope you upgrade; $5/month Solo and $15/month Pro are the flat, meter-free steps up.
One boundary is worth stating plainly, because it is the only one: Free is for development, testing, evaluation, CI, personal projects, and early-stage products, and once a product is running in production and earning recurring revenue it belongs on a paid plan. There is no clock on that and nothing enforces it automatically. If your usage makes it obvious, you get a friendly email asking you to move up, never a stopped database. The full clause is what the Free plan is for.
But the free tier is not a trial with the label filed off, and it is not bait for a future repricing. It is the part of the product the architecture makes cheap to keep, which means it gets to keep existing.
FAQ
Is the Layerbase free tier really free forever?
There is no card, no expiry, and no trial clock. The reason it can stay is economic rather than charitable: a sleeping database holds no RAM and no CPU, so an abandoned side project costs us a few gigabytes of storage and nothing else. We are not promising generosity against our own interests, which is the only kind of free-tier promise worth trusting.
Do free databases sleep, and how long do they take to wake?
They sleep after 15 idle minutes and come back in a few seconds on the next connection. There is no dashboard button and no restore flow, so your client sees a slightly slow first connection rather than an error. A keep-alive cron job is not needed and would only defeat the point.
What happens if I ignore a free database for months?
Nothing you cannot undo. It sleeps, it keeps its data, and it wakes when you reconnect. The guarantee underneath is that we never delete your data out from under you: every destructive path takes a backup first, deletion is refused if a final backup cannot be secured, and the most recent backup is never pruned. That applies to free databases, not just paying ones.
Which engines can I run on the free tier?
Eight: PostgreSQL, MariaDB, SQLite, DuckDB, libSQL, Redis, Valkey, and TypeDB. A relational database and a cache, the standard two-piece stack, fits inside the free tier with room left over. The heavier engines, ClickHouse, QuestDB, Qdrant, Weaviate, and InfluxDB, need more resources than a free slot provides and live on the Pro plan.
Why do database free tiers keep getting cancelled?
Because a conventional managed database is a process that runs all the time and holds RAM whether anyone queries it or not. Multiply that by a few hundred thousand mostly-abandoned free databases and the tier becomes a permanent tax on the vendor. The usual defenses, pausing after a week or capping at 30 MB or replacing the tier with a trial, are all versions of making idle databases cheaper by making them worse.
What does upgrading get me?
Paid plans can pin databases always-on, so the first query after an idle hour is not slow, and they open up the heavier engines. Solo is $5/month and Pro is $15/month, both flat and meter-free.
Starting
Create a free database. Two databases, eight engines, 5 GB, no card. Come back in six months and it will still be there, asleep, waiting for a connection.
Keep reading
- Serverless database pricing compared: what each meter actually countsNeon counts compute units. Aurora counts ACU-hours. Cloudflare D1 counts rows scanned. Upstash counts commands. Turso counts rows read. Every serverless database bill is a formula, and the variable is always something your code controls without telling you. Here is each meter, and what a flat price costs instead.
- What Is a Serverless Database?Serverless does not mean there is no server, and it does not mean cheap. It means nobody hands you a capacity slider, and an idle database should not cost what a busy one costs. Here is what the term actually promises, and the two very different ways vendors deliver it.
- Every Free Database Tier That Sleeps, Pauses, or Expires - and What Staying Awake Actually CostsA database that scale-to-zeros after five minutes, a project that pauses after a week and needs a human to click Resume, and a database that gets deleted 44 days after you created it are three different products. Here is which vendor does which, and what the cheapest always-on version costs.
- Fly.io is raising Machine memory prices 20% on October 1Fly emailed customers that Fly Machines memory prices rise 20% on October 1, 2026. CPU stays the same, and Sprites get cheaper. Here is what that does to a Postgres cluster, and how to move the database without moving the app.