Everpure used its Pure//Accelerate event in London to expand the Data Intelligence platform it launched in June, adding a lightweight file-based edition, a native Model Context Protocol (MCP) server, and a set of Unified Data Plane enhancements the company says can cut stored data volumes roughly in half and speed AI inference time-to-first-token by up to 20x.

The announcements build on the "data-primacy" architecture Everpure introduced at its Las Vegas conference in June. That launch, Everpure Data Intelligence, rests on data discovery and classification technology the company acquired with 1touch.io in early 2026. The product scans metadata across Everpure storage and third-party platforms and builds a semantic knowledge graph of a customer's entire data estate.

The premise is straightforward: many AI project failures are data management failures in disguise. AI depends on data quality for both training and inference, and gaining visibility into quality, availability, security, privacy, and provenance has defeated large cohorts of big data, business intelligence, and data lake initiatives. Data changes constantly, so perfect data management remains out of reach — but Everpure is betting there is plenty of room for improvement.

Two-tier classification

The first of today's additions is Data Intelligence File Edition. The existing product, now renamed Data Intelligence Context Edition, applies the full stack of discovery, classification, and contextualization techniques, and builds the knowledge graph that relates data items to each other. File Edition takes a lighter approach.

"Context Edition is the full product which has discovery classifier and contextualization. It comes with the knowledge graph that starts to build relationships between the data," said Ashish Gupta, general manager of the Everpure Data Management business unit. "File Edition is essentially taking the physical layer of the file, looking at all the metadata around that file, and classifying the file as being sensitive or not sensitive."

The rationale is resource economics. Many estates hold large volumes of ROT data — redundant, obsolete, trivial — and running deep classification over it wastes time and compute, Gupta said.

Once data passes through either classification tier, customers can query it through a new native MCP server embedded directly in Data Intelligence. Gupta framed the capability in compliance terms: "you can say 'Tell me where my GDPR data is sitting' and 'Tell me which areas are not in compliance with the GDPR rules.'" The mapping extends beyond Everpure arrays to cloud storage, other on-premises systems, and even mainframe environments, he said.

Unified Data Plane gains

Everpure also upgraded its Unified Data Plane, the software layer that manages file, block, and object data across Everpure and third-party storage.

DeepReduce, the similarity-based reduction technology that operates on already-compressed data stored on Everpure hardware, now delivers about a two-to-one reduction, according to Vishwajeet Mishra, Everpure director of product marketing. "You get that without some of the performance overhead," he said.

On the AI side, enhancements to the Key-Value Accelerator (KVA) pre-stage context directly into GPU memory, which Everpure says delivers up to 20x faster time to first token. "You get better, faster inferencing results, more inferencing per GPU," Mishra said.

Evergreen//One, the storage-as-a-service subscription spanning on-premises, colocation, and public cloud, received new service level agreements. "You double the performance on the minimum guaranteed performance for all three tiers," Mishra said.

Two reference architectures round out the Data Plane news. A token optimization architecture based on open weight models aims to give customers more control over their data and more predictable costs from external AI providers. Additional architectures for Red Hat AI Factory adopters — one available at launch, two to follow — were built with Red Hat and Nvidia. Everpure also added Red Hat OpenShift support on Azure, flexible storage for AI workloads via Portworx, and improved file protection and performance controls.

Control Plane updates

The Intelligent Control Plane, which manages storage fleets regardless of vendor, gained a beta of the Pure1 Configuration Advisor. A typical customer runs hundreds of distinct, individually managed storage configurations; the advisor consolidates them into a small set of policy presets.

"Now imagine a scenario where you can consolidate these hundreds of configs based on specific requirements," Gupta said. "That might be your security requirement or your QoS requirements, [consolidated] into a set of five or six policy groups or presets that really meet most of your workload profile." Configurations can be migrated non-destructively for penetration or performance testing, reducing management overhead.

Everpure also disclosed a beta automation workflow that feeds Data Intelligence insights into storage management through Fusion, its self-service platform covering Everpure and non-Everpure storage, with a general announcement expected in the second half of the year. Portworx, Everpure's Kubernetes environment, gained observability capabilities.

With File Edition lowering the entry cost of classification and MCP integration making data estates queryable by AI agents, Everpure is positioning storage management as a prerequisite for AI ROI — a bet that will be tested as the remaining Data Plane and automation roadmap items ship through the end of the year.