MariaDB Has Vector Search. Coding Assistants Keep Recommending Someone Else's Database

MariaDB Community Server has shipped native vector search since version 11.7, generally available in February 2025, with the 11.8 LTS release following in May 2025 as the recommended production version. This is not a preview feature or a bolt-on plugin. The project's own documentation for AI agents states plainly that the capability is built in: "MariaDB has native built-in vector support since version 11.7; it is NOT a plugin or extension (unlike pgvector for PostgreSQL)," according to the SKILL.md file maintained in the MariaDB Foundation's public skills repository.

Despite that, MariaDB Foundation executive chairman Kaj Arnö told The Register in a September 2026 interview that AI coding assistants keep pointing developers elsewhere when they ask for help building applications with vector search. His complaint, reported by The Register, is not that MariaDB lacks the feature. It is that when a developer types a prompt into an assistant like Claude Code and lets the tool pick a database, the assistant's default suggestion rarely comes back MariaDB.

Arnö's framing draws a direct line to an earlier era of software marketing. He compared the situation to the days of chasing Google search rankings, calling that old problem "a mystical science," according to reporting from Windows Forum. The comparison lands because the mechanics are similar: a system that developers trust to make a recommendation, working from opaque signals nobody outside the vendor fully controls, now decides which product gets tried first.

His diagnosis of the causes is his own theory, not a measured finding, and he says as much. Training data may simply predate MariaDB's vector rollout for some models. But the more persistent gap, in his account, sits in the layer between the database and the developer: framework connectors, tutorials, and documentation written in a form that coding assistants can actually ingest. Semantic Kernel and Haystack, two widely used frameworks for building AI applications, do not yet ship MariaDB connectors, according to the Windows Forum account of the interview. Until that changes, a developer who wants MariaDB Vector inside those frameworks has to override the assistant's default and ask for it by name.

MariaDB's response has been to build the machine-readable layer directly rather than wait for third parties to write it. The Vector Overview documentation points to an index file intended for the kind of automated retrieval that agent tools perform, and the project has published the SKILL.md file specifically so that coding assistants have something structured to read when a user asks about vector search in MariaDB. It is an attempt to feed the same pipeline that currently favors competitors.

Arnö has ruled out the alternative approach of gaming assistant recommendations the way sites once gamed search rankings. "The only proper way to influence them is by having it and telling people," he said, per the Windows Forum writeup of his Register interview. That leaves MariaDB dependent on the same slow mechanism that used to govern search visibility: get the documentation right, get the feature genuinely working, and wait for the tools to catch up to what already shipped over a year ago.