Quesma has built a gateway between Elasticsearch EQL and SQL-based databases like ClickHouse, claiming EQL users can use it to access faster and cheaper stored data sources.
EQL (Elastic Query Language) is used by tools such as Kibana, Logstash, and Beats. Structured Query Language (SQL) is the 50-year-old standard for accessing relational databases. Quesma co-founder Jacek Migdal, who previously worked at Sumo Logic, says that Elasticsearch is designed for Google-style searches, but 65 percent of the use cases come from observability and security, rather than website search. The majority of telcos have big Elastic installations. However, Elastic is 20x slower at answering queries than the SQL-accessed ClickHouse relational database.
Quesma lets users carry on using Elastic as a front end while translating EQL requests to SQL using a dictionary generated by an AI model. Migdal and Pawel Brzoska founded Quesma in Warsaw, Poland, in 2023, and raised €2.1 million ($2.3 million) in pre-seed funding at the end of that year.
The company partnered with streaming log data lake company Hydrolix in October 2024 as it produces a ClickHouse-compatible data lake. Quesma lets Hydrolix customers continue using EQL-based queries, redirecting them to the SQL used by ClickHouse. Its software acts as a transparent proxy.
Hydrolix now has a Kibana compatibility feature powered by Quesma’s smart translation technology. It enables Kibana customers to connect their user interface to the Hydrolix cloud and its ClickHouse data store. This means Elasticsearch customers can migrate to newer SQL databases while continuing to use their Elastic UI.
Quesma enables customers to avoid difficult and costly all-in-one database migrations and do gradual migrations instead, separating the front-end access from the back-end database. Migdal told an IT Press Tour briefing audience: “We are using AI internally to develop rules to translate Elasticsearch storage rules to ClickHouse [and other] rules. AI produces the dictionary. We use two databases concurrently to verify rule development.”
Although AI is used to produce the dictionary, it is not used, in the inference sense, at run time by customers. Migdal said: “Customers won’t use AI inferencing at run time in converting database interface languages. They don’t want AI there. Their systems may not be connected to the internet.”
Its roadmap has a project to add pipe syntax extensions to SQL, so that the SQL operator syntax order matches the semantic evaluation order, making it easier to understand:
Quesma is also using its AI large language model experience to produce a charting app, interpreting natural language prompts, such as ”Plot top 10 languages, split by native and second language speakers” to create and send requests to apps like Tableau.
