From Edge Vector Indexing to LLM Models: What’s New in Memgraph 3.4

At Memgraph, we’re always working to make your experience with graph databases smoother, faster, and more powerful, especially when it comes to leveraging the full potential of large language models.

That's why we're excited to announce the release of Memgraph 3.4.0, MAGE 3.4.0, and Lab 3.4.0.

This release brings new features and performance improvements that give you more control and flexibility for building intelligent, context-aware LLM applications.

Let’s dive into what’s new!

Vector Indexing Gets Smarter (and Smaller)

Vector search isn’t just for nodes anymore.

You can now create vector indexes on edges, allowing for similarity searches on relationships. This is especially useful when working with embeddings generated for interactions, like recommendation engines, fraud detection, etc.

CREATE VECTOR EDGE INDEX edge_index_name
ON :INTERACTS(embedding)
WITH CONFIG {
  "dimensions": 128,
  "scalar_kind": "f16"
};

But that’s not all! Need to shrink memory usage? You can now apply quantization when building a vector index. Just tweak the scalar_kind in your config map to reduce footprint. This is useful when working at scale or on memory-constrained deployments.

Tradeoff: You’ll save memory at the cost of some precision. But if your use case can tolerate that, it’s a simple win.

Monitoring Replica Recovery

For those running replicated setups, we've added monitoring of the replica recovery process. You'll now be able to track:

• How often a replica enters recovery state
• How often it recovers successfully

This gives you better visibility into the health and resilience of your replicated Memgraph instances.

Count Your Hops

Need to know how many hops your graph traversal has made?

The new built-in function getHopsCounter() returns the current number of hops traversed in a query. That’s useful for debugging complex traversals or or implementing hop-count-sensitive logic directly in your queries.

Index Creation, Now Non-Blocking

Creating indexes used to lock the database for writes.

Not anymore.

Memgraph v3.4.0 introduces non-blocking index creation. It waits for in-flight writes to finish, queues new ones briefly, then builds the index in the background. All this will be done while allowing reads to continue.

The result: shorter pauses, smoother deployments, and no risk of data loss even under heavy load.

GraphQL Compatibility Update

We’ve added support for @neo4j/graphql middleware v7.2.0.

That means better integration with the GraphQL ecosystem, fewer issues when migrating apps, and an easier time building modern APIs on top of Memgraph.

MAGE v3.4.0: Expanding Your Graph Analytics Toolkit

The latest MAGE update brings a set of small but powerful utilities:

• text.replace() to find and replace strings.
• text.regreplace() to use regular expressions for more complex substitutions.
• text.distance() to calculate the distance between two text values.
• date.convert_format() to convert temporal strings from one format to another.
• refactor.mergeNodes() to combine duplicate or redundant nodes based on shared properties.

These additions simplify cleaning and transforming your graph for better graph data manipulation and analysis.

Memgraph Lab v3.4.0: More Views, More Control

Memgraph Lab just got a couple of major usability upgrades:

• New Graph Layouts: Choose horizontal or vertical hierarchical views. These are perfect for tree-like structures and layered graphs.
• Graph Chat LLM Options: Now you can plug in Anthropic and Gemini models for Graph Chat, expanding your model options and giving you more flexibility and customization when building LLM applications.

• Security Boost: Lab now supports PKCE (Proof Key for Code Exchange) for OIDC SSO, significantly improving security in OAuth2 authentication workflows.

Take It for a Spin

From how you store embeddings to how you visualize your graphs, this update gives you tighter control over performance, better indexing, and new flexibility across your entire Memgraph stack. We're particularly excited about the advancements in vector indexing and LLM integration, as they lay the groundwork for even more intelligent graph solutions.

Download Memgraph 3.4.0, MAGE 3.4.0, and Lab 3.4.0 today and start exploring these new features. We're confident you'll find them invaluable for your next project.

Have questions? Hop on Discord or explore the Docs.

We’re excited to learn about what you’re building!