The Free Tier That Stays

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 90 days, 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).

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 60 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. 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 60 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 an idle hour 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.

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.

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.