DocsConcepts

Ephemeral databases

How seedkit's ephemeral Postgres works under the hood — Neon, scale-to-zero, TTL, lifecycle, restore.

When you run seedkit new, we provision a fresh Neon project — a real Postgres 16 instance with its own URL. This page covers what that means in practice.

Where they run

Provider: Neon. Postgres 16. Serverless architecture.
Region: EU (eu-central-1) by default. Override with --region. Available: eu-central-1, us-east-1, us-west-2, ap-southeast-1.
Compute: scales to zero between queries. First query after a sleep wakes the DB in ~200ms.

This is what makes the free tier sustainable — your DB doesn't burn compute when nobody's hitting it.

Lifecycle

Phase	Trigger	What happens
Provisioning	`seedkit new`	New Neon project + branch + role created. Schema applied. Cached SQL replayed (or freshly generated).
Ready	provisioning ✓	Connection string is live. Queries work.
Sleeping	no queries for ~5min	Compute scales to zero. Connection still works; first query wakes it.
Expired	TTL elapsed	Neon project + connection deleted. Cached SQL survives.
Revived	`seedkit revive`	Fresh Neon project; cached SQL replayed. New connection string.

TTL by tier

Tier	Default TTL	Max TTL	Concurrent DBs
Free	1 hour	1 hour	1
Pro	24 hours	7 days	5
Team	7 days	30 days	25

Set with --ttl 6h or --ttl 3d on seedkit new / seedkit revive.

What's persisted vs. destroyed

Persisted forever (until you --purge or delete via the dashboard):
- The schema spec
- The cached SQL dump (your seed)
- The seed name → key mapping
Destroyed on TTL or seedkit destroy:
- The Neon project
- Compute credentials and connection string
- Any data you wrote outside of the seedkit insert (e.g. rows you INSERT-ed manually after the fact)

Restore

Run seedkit revive <name>. We provision a fresh Neon project and replay the cached SQL. New URL, identical data.

seedkit revive crm-demo --ttl 24h --env

See seedkit revive for the full reference.