databases

A real Postgres for every branch,
warm before the first query.

Every database is its own Firecracker microVM running PostgreSQL 16 — its own kernel, its own postgres, a connection pooler, and a durable volume. Branch it while it runs, rewind it to any moment, and pay nothing for compute while it sleeps.

1 VM
per database
$0
compute while idle
90 days
PITR window on Team
postgres 16 — branch + point-in-time restore
isolation

Your database is a computer, not a tenant.

Shared-cluster serverless Postgres hands you a slice of someone else's machinery. A PandaStack database is a dedicated microVM: its own kernel behind KVM, its own postgres process, PgBouncer in front, and a durable data volume underneath — reachable by a native postgres:// URL in one API call.

Why the whole machine matters

Databases are the guaranteed workload class: memory is never overcommitted, never squeezed, and the TTL reaper that recycles ordinary sandboxes never touches them. Data lives on a dedicated durable volume — it survives hibernate, wake, and host reboot. pgvector, pg_trgm, and pgcrypto come pre-installed.

own kernelown postgresdurable volume
python — pip install pandastack
import pandastack

client = pandastack.Client(api_key="pds_...")
db = client.databases.create(label="my-app-db")
print(db["connection_url"])
# postgres://pandastack:<pw>@<id>.db.pandastack.ai:5432/pandastack

# "undo" a bad migration: a NEW database as of 09:30, source untouched
clone = client.databases.clone(
    db["id"],
    label="before-the-bad-migration",
    target_time="2026-08-06T09:30:00Z",
)
branch

Branch a running database, keep the cache.

A warm branch is a new database born from a live one — its first query hits a hot cache instead of cold storage, so there is no ramp-up while it re-reads what the parent already knew. Each branch diverges copy-on-write; the parent never notices.

Warm from query one

The branch arrives with the parent's working set already hot. Point a preview, a test suite, or an agent at it and the first read is fast — no cold-cache tax.

Its own front door

Every branch gets its own connection string, fresh credentials, and an isolated backup stream. Wiring a throwaway environment never risks production.

Copy-on-write divergence

Branches share data with the parent until they write, then diverge page by page. Keep the branch that worked, delete the rest — the source stays untouched.

recover

Any moment in the window is restorable.

Daily base backups plus continuous WAL archiving — automatic, nothing to enable — give point-in-time recovery to any instant in your retention window: 7 days on Free, 30 on Pro, 90 on Team and Enterprise. Not just the moments a backup happened to run.

Restore in place

Roll this database back to a point in time. Id, host, and connection string — password included — are preserved, and a safety backup of the current state is taken first, so the restore itself is undoable.

Clone

Provision a brand-new database from this one's backups, optionally at a point in time. The source keeps running unaffected — and a clone can change RAM tier, which is the supported resize path.

Failover

If the host is lost, rebuild the database on a healthy machine from the off-host archive. Same id, same host name; the dashboard offers it automatically when a database reports failed.

connect

Connect like it's Postgres. Pay like it's serverless.

A native postgres:// URL that works with every driver and ORM, pooling built in, an HTTP broker for the places TCP can't go — and compute that stops billing the moment nobody is connected.

TLS, always

Traffic is routed to your VM by SNI and TLS is required — sslmode=require with any PostgreSQL client.

Pooled by default

PgBouncer in transaction mode accepts up to 500 client connections and multiplexes them — built for serverless and short-lived agent connections.

REST query broker

Query, exec, and transactions over HTTPS with a scoped broker token — for edge functions and anywhere a raw TCP connection can't open.

Scale-to-zero

An idle database pauses its compute and bills $0; the next connection wakes it in seconds. Set always_on for workloads that must never wait.

compare

The difference is the machine.

Serverless Postgres re-architects the database around shared storage. PandaStack keeps stock PostgreSQL 16 and gives it a whole microVM instead.

PandaStack databaseshared-cluster serverless Postgres
Isolationdedicated kernel per database (KVM)tenants separated inside shared machinery
Noisy neighboursnone — the VM is yoursshared compute and shared storage tiers
Branch cachewarm — inherited from the parentcold — re-read from remote storage
Readslocal durable volumenetwork round-trip to a pageserver
Apache-2.0 · self-host or managed cloud

Ship on the millisecond cloud.

Free tier with $5.40/mo usage credit. No card. Apache-2.0.