Elastic Unveils AI-Era Metrics Platform, Claims Major Speed and Storage Gains

Elastic outlined its new metrics offering during an observability-focused investor call, positioning the product as a faster and more storage-efficient way for customers to manage metrics alongside logs, traces and other data types on a unified platform.

Baha Azarmi, Elastic’s General Manager of Observability, said the offering launched near the start of the company’s fiscal year in June after the company re-architected Elasticsearch for metrics workloads. The product is designed to address rising data volumes associated with artificial intelligence deployments, including monitoring of large language model calls, graphics-processing-unit activity, retrieval-augmented-generation queries and AI agents.

“AI is really causing an explosion in metrics,” Azarmi said, describing how agent-based workloads can create less predictable telemetry volumes than conventional applications. He said organizations may face blind spots if cost pressures force them to limit the metrics they retain.

Elastic said it built a column store within Elasticsearch specifically for metrics, which typically consist of numerical values, timestamps and multiple dimensional labels. Azarmi said the approach is intended to accommodate high-cardinality data, meaning data with a large number of unique dimensional combinations, without restricting customers’ use of dimensions.

The company said it uses techniques including dimension filtering and storage-codec tuning to improve query and storage efficiency. Azarmi said Elastic has published benchmarks and open-source code on GitHub that compare the offering with other products. According to the company’s benchmarks, Elastic’s metrics solution is 30 times faster than Prometheus and Mimir and eight times faster than ClickHouse. Those performance claims were presented by Elastic and were not independently verified on the call.

Azarmi said the platform combines a document store for logs and traces, a column store for metrics, and a vector store for vectorized data. The integrated approach is intended to allow users and AI agents to investigate issues across metrics, traces, logs and knowledge bases from within the same platform. He also said Elastic recently acquired Deductive AI, a company focused on root-cause-analysis investigations with agents.

Elastic plans to share additional observability-related announcements at its ElasticON event in New York on Oct. 8.

Elastic emphasized compatibility with PromQL, the query language widely used with Prometheus-based metrics systems. Azarmi said users can bring PromQL queries into Kibana, and that Elastic has reached 90% compatibility with popular PromQL dashboards while working toward full compatibility. The company also highlighted migration tools intended to help customers move from existing metrics platforms, along with professional-services support. Its out-of-the-box experiences include dashboards, visualizations, alerts, service-level objectives, machine-learning jobs and workflows, beginning with Kubernetes and AWS. Elastic also cited technology-specific integrations for Temporal, Supabase and Vercel, including managed endpoints for ingesting metrics.

Users can access metrics through Elastic’s user interface, natural-language queries via its AI agent, or through tools including model context protocol servers, tools, skills and applications, Azarmi said.

Elastic identified three primary sales opportunities for the offering: Adding metric...