MariaDB starts a database platform for AI agents

MariaDB, a company that makes databases, started Enterprise Platform 2027, and customers can use it at this time. The platform puts agent memory, a cache, live analytics and vector search, which finds data by meaning, in one system. CEO Rohit de Souza tells that a model is not sufficient, and agents must have one data system below them.

What is new

• Agent Memory: data that agents keep between sessions, to continue their work.
• Semantic Cache: keeps the result of a query. A query with the same meaning gets that result.
• AI Retrieval: changes text from a person into a database query that only reads.

The numbers

MariaDB tells that an external test found 56% more peak throughput than the last primary release. Its vector search gives a maximum of 30% more queries for each second. The source does not name who did the test and gives no test conditions.

What is not known

MariaDB tells that one system gives agents less latency, the wait for a result. It tells that each connection adds latency and a point of failure. It gave no test of agent results, and AI Retrieval is only in public preview on MariaDB Cloud.

Insight: Three vendors claim the layer below agents

MariaDB wants all agent data in one database. DataHub makes a layer that tells agents which data is right. Gartner at this time makes a report of platforms that operate agents. This reads as a plan by each vendor to be the layer below the agents. Monitor if buyers select one platform or keep some.