Aura Graph Analytics - Serverless on a Self Managed Neo4j Instance

Aura Graph Analytics is an on-demand ephemeral compute environment for running GDS workloads. Graph projections and algorithms are run in a separate dedicated Aura environment, thus do not consume the resources of the environment hosting the Neo4j database. The GDS version running in that environment will always be the latest available, which will always require a recent version of Neo4j.  Currently provided as of this writing is Aura GDS version 5.26 which requires Neo4j 2026.01.3. Refer to the GDS Support Neo4j Versions matrix to determine the minimum required Neo4j version for Aura Graph Analytics Serverless. Graph Analytics Serverless requires a subscription, and an API Key generated in the Aura Console.

Graph Analytics Configurations

There are four intended configurations for GDS: using the plugin (embedded), connected to an Aura instance (Attached), connected to a self-managed instance, or standalone. 

Embedded

Adding the GDS Plugin to a Neo4j instance, and using GDS via the plugin is the original intended application of GDS, and thus the most reliable. However, all projections and algorithms use the same resources as the Neo4j database, which can cause significant performance issues for the database. This has led to many production environments cloning into separate environments dedicated to GDS, which itself led Neo4j to implement the serverless options that are the subject of this article. 

Attached

Establishing a GDSSession connection to an existing Aura instance to project data into an Aura environment dedicated to GDS. Graph Analytics Serverless is primarily tested against Aura environments. All tiers of Aura are supported. 

Self-Managed

Establishing a GDSSession connection to a self-managed Neo4j instance brings the same benefits as running attached to an Aura instance, but there are known issues with write-back, and requires a recent version of Neo4j. (LTS versions will not work with Graph Analytics Serverless)

Standalone

Opening a connection to the GDS Aura environment without connecting to an existing database. Standalone Graph Analytics allows loading many other datasources, typically using Pandas Dataframes as an intermediary to build a projected graph. 

PrivateLink for Self-Managed

Connecting to a self-managed Neo4j Enterprise instance in a private VDC can be done either by opening ports (7687, 8491) to public, or via PrivateLink. Connecting via PrivateLink requires a few additional configurations to the environment in the VDC:

• Neo4j instance must be recent. (latest from the GDS support versions matrix)
• Add and configure the GDS Plugin.
• Arrow must be configured. ( gds.arrow.enabled=true )
  • Set listen address if instance is behind a load-balancer. ( gds.arrow.advertised_listen_address )
• Route 53 CNAME wildcard pointing to the Aura orchestration endpoint. 
  •  *.internalname 3600 CNAME {dbid}.production-orch-1234.neo4j.io 
• Connect via  neo4j://   not   bolt:// 

Known Issues

Write-back
Neo4j Engineering team is actively working on expanding GDS write-back functionality for Attached and Self-Managed implementations, but most of the write-back methods are non-functional as of this writing.

Known Errors

gds.graph.project doesn’t exist
incorrect connection string, or connection credentials to the Neo4j instance for the connection being passed to GDSSession.

There is no procedure with the name `gds.arrow.project` registered for this database instance.
Connecting to an older version of Neo4j which is not supported by the current GDS.

The procedure or function gds.session.dbms.protocol.version() was not registered for this database instance.
Connecting to an older version of Neo4j which is not supported by the current GDS.