📢 Memgraph 3.14: Cypher 25 subqueries, a production-ready disk engine, and a FIPS image

Next Hop

One hop further into graphs

and AI context, every Wednesday

Hey there,

Memgraph 3.14 is out, and if you’re running Memgraph in production with heavy writes or on disk, this is the upgrade you’ve been waiting for.

What’s new in Memgraph 3.14

• Cypher 25 subquery additions: conditional CALL, OPTIONAL CALL, aggregation in CASE, and label expressions
• Sort and comparability correctness: a comprehensive pass to bring ORDER BY and comparison operators into full openCypher spec compliance
• Disk engine stabilisation: around 40 fixes that make the disk storage engine production-ready
• Engine-lock throughput: adaptive admission that eliminates BEGIN/PREPARE blocking under write-heavy load
• Auth hardening in Enterprise: transactional auth, session privilege revocation on user/role drop, and rolling-upgrade auth safety
• A FIPS-compliant image for regulated industries and large enterprises

Waypoints

Your coding agent forgets everything between sessions, including the decision you made, the fix that worked, and the workflow you explained twice.

Ante Javor shows how to give it a memory that sticks, using a reference implementation from Memgraph’s AI Toolkit that installs in one line for Claude Code or Codex. Everything your agent knows, did, and has learned to do lives in one graph you can open and query with Cypher.

Jacob Coles from Redfield shows how investigators find suspicious payment patterns across hundreds of documents. LLMs extract the entities and relationships into Memgraph, and everything runs locally.

Alen Zubic explains how RelayMesh keeps mesh telemetry in ClickHouse and builds short-lived Memgraph graphs to answer topology questions, such as where a new relay would help.

Routing Table

• You Built a Knowledge Graph From 1 Million Documents. Can You Trust It?
  Building the graph is expensive. Finding the wrong edges afterwards is harder, so this post covers how to plan for it.
• The Single-Writer Bottleneck
  Here's why engines that run one write at a time per graph hit a ceiling when fraud, security, or agent events keep updating the same graph.
• Why Retrofitting Real-Time Graph Workloads Gets Expensive
  Learn how storage, concurrency, and execution models set limits that a cache or an extra service can’t remove later.