For MariaDB, LLMs are the new search engines to win over

MariaDB has integrated vector search capabilities into its database technology. However, company executives face a challenge in convincing artificial intelligence coding tools to highlight this feature, according to Kaj Arnö, the chairman of the MariaDB Foundation. At Percona Live, Arnö expressed frustration that the world remains largely unaware of MariaDB's vector search functionalities.

He noted that rival database vendors seem to be in a more collaborative atmosphere, but he acknowledged concerns about MariaDB's competitiveness against MySQL, which was forked by MariaDB in 2009 following Oracle's acquisition of Sun Microsystems.

Vector search allows applications to find items based on similarity rather than exact matches, a requirement frequently encountered in AI applications. The MariaDB Foundation supports the open-source development of the server and its community, while MariaDB plc handles the commercial products and provides support. Arnö expressed worry that developers embarking on new AI projects may not be directed towards MariaDB for vector search.

He explained that many developers rely on frameworks or default choices suggested by large language models (LLMs) when beginning AI projects. If an AI tool such as Claude Code recommends a database during coding, users may not end up using MariaDB for vector search purposes.

Part of the problem lies in the information available to AI tools. Their training data may not have included MariaDB's vector search features, and the documentation and examples developers find elsewhere might still promote other solutions. Arnö emphasized the need to update these resources so that frameworks, training data for LLMs, and other materials acknowledge the existence of MariaDB vectors.

He illustrated this with a comparison to historical search engine optimization (SEO) practices, where understanding search engine algorithms was less transparent.

To influence AI tool recommendations, Arnö believes the proper approach is through documentation, usage, and community awareness. However, MariaDB faces a chicken-and-egg dilemma: without users generating examples and discussions of its vector search capabilities, AI tools have limited material to learn from. Early adopters may gain an advantage, similar to how certain developments initially capture market momentum.

MariaDB needs to change this trajectory and raise awareness about its vector search capabilities to challenge the status quo.