# A graph database that packs light

Your application has enough to carry without its database arriving with a
removal van. Galactus DB ships as a small, self-contained image, with Explorer
in a separate package when you want a visual workbench.

That makes for a quick download and leaves more room for the interesting
part: your graph. Here is what was in the suitcase in our September snapshot.

## Small enough to notice

These figures compare the Galactus images and official `neo4j:5` Community
image available on 17 September 2026. Tags move, so the numbers belong to
those builds.

| Image | Uncompressed | Download (amd64) | Download (arm64) | Layers | Idle memory after 60 s |
|---|---|---|---|---|---|
| `ianknowles/galactus-db:latest` | 5.9 MB | 1.9 MB | 1.7 MB | 2 | 8.1 MB |
| `ianknowles/galactus-db-explorer:latest` | 1.7 MB | 0.8 MB | 0.8 MB | 1 | 1.9 MB |
| `neo4j:5` (Community) | 627.2 MB | 372.6 MB | 370.7 MB | 5 | 534.7 MB |

The Galactus database and Explorer together downloaded in under **3 MB** and
used about **10 MB of RAM** while idle. The Neo4j database image was roughly
**105 times** larger on disk, **193 times** larger to download on AMD64, and
used about **66 times** the idle memory of the Galactus database image.

Those packages bring different things with them. Neo4j includes a Java
runtime, Browser, `cypher-shell`, plugin loading and a Debian base. Galactus
packages the database and workbench separately, which helps explain the gap.

![Image size: uncompressed and download size](/blog/charts/image-size.svg)

![Idle memory after 60 seconds](/blog/charts/idle-memory.svg)

## What did we leave out of the suitcase?

The database ships as a statically linked Rust binary in a `scratch`
container. There is no operating-system layer, shell or separate language
runtime to download. The image contains the binary and a data directory
ready for use.

Explorer is optional. If your application talks directly to the database,
you can leave the workbench image on the shelf. When you do want to explore
connections visually, it is there to pull separately.

The database also runs as an unprivileged user and includes its own health
probe, so it does not need a shell just to tell your container platform that
it is listening.

## Where packing light pays off

Think of a fresh development machine, a CI job starting from an empty image
cache, or a deployment over a modest connection. A smaller download gives
each less to fetch before the application can get on with its work.

Low idle memory also leaves breathing room for the application. It is useful
when you run several environments, or simply prefer your database to settle
in without taking over the sofa.

There are fewer bundled components to maintain, too. You still need to keep
the database up to date and configure access for your deployment; a small
container needs looking after just like a large one.

## A small image can hold a growing graph

Download size and data capacity are different questions. Nodes, relationships
and indexes need space as you add them. Galactus keeps graph structure and
indexes in memory even when base properties are stored on disk, so size your
instance for the graph you intend to run.

For memory use with data and queries in the mix, head to the
[Neo4j comparison](/blog/galactus-db-vs-neo4j).

## About the figures

The comparison used each image's defaults with Docker Desktop 28.0.1 on
Windows 11. Uncompressed size and layers come from `docker image inspect`;
download sizes are the compressed image sizes reported by Docker Hub for
each architecture. Idle memory is the `docker stats` reading after 60 seconds
with no connections.

Neo4j's heap and page-cache settings affect its memory use. These are readings
for the packaged defaults, rather than minimum requirements for either
database. They are a guide to what arrives out of the box.

Neo4j is a registered trademark of Neo4j, Inc. Galactus DB is not affiliated
with Neo4j, Inc.
