AgensGraph 2.18 moves to PostgreSQL 18.6 and adds the first ISO GQL clauses next to openCypher
The AgensGraph team has released version 2.18.6.1, the latest build in the 2.18 series of the open-source graph database that runs on PostgreSQL and lets SQL and graph queries work on the same data in the same transaction, even in one statement. The project has been developed since 2016 by SKAI Worldwide, formerly Bitnine, under the Apache 2.0 licence.
The language change is a first step towards ISO/IEC 39075, the international GQL standard for property graphs. Release 2.18 adds the LET, FOR, FILTER, NEXT, FINISH and INSERT clauses, standalone ORDER BY, OFFSET and LIMIT, and graph subqueries such as CALL, OPTIONAL CALL, EXISTS, COUNT, COLLECT and VALUE, plus CALL with YIELD to invoke PostgreSQL functions from Cypher. The team says more GQL clauses will follow in later releases.
Speed was the other theme. Variable-length traversals now use the edge label's index at each hop instead of scanning the whole label, and on the public soc-Pokec graph with 30 million edges a two-hop traversal that had not finished within 20 seconds on 2.17 now returns in milliseconds. Frequently queried properties can be stored as typed columns and indexed like ordinary PostgreSQL columns. Cypher writes now respect row-level security.
The base moves to PostgreSQL 18.6, bringing asynchronous I/O, uuidv7 and skip scans. A 2.17 data directory cannot be reused directly: upgrades go through pg_upgrade or a dump and restore with the 2.18 pg_dump. A new Python driver, agensgraph-python 2.1, is rebuilt on psycopg 3.

What it means
GQL support matters because it gives graph queries a standard that is not tied to one vendor. Users with large traversals should test the index change first: it is the difference between a query that times out and one that returns.