London – – September 30, 2026 -- Everpure (NYSE: P) has launched new platform capabilities that cut large language model inference latency by up to 20x and give enterprises native access to governed data for AI agents, the storage and data management company announced. The updates, rolling out in October, extend the company's Data Primacy architecture first introduced at Pure//Accelerate in June.
PureKVA accelerator slashes AI response times by up to 20x
Everpure's FlashBlade system now pre-stages context directly into GPU memory through a new Key-Value Accelerator, delivering up to 20x faster Time to First Token. The company said the approach supports multi-tenancy without relocating datasets, eliminating GPU idle time and reducing response lag for real-time applications.
Native MCP integration removes custom API work for AI agents
Everpure Data Intelligence now implements the open Model Context Protocol, allowing AI agents and security tools to query live enterprise data catalogs using natural language. The system discovers, classifies, and contextualizes information across the Everpure Platform, public clouds, SaaS applications, and third-party storage, letting agents assess data sensitivity before use.
Privacy-first file intelligence flags exposure risk before AI access
A new file intelligence feature identifies who can access each file share and how stale the data is, without reading file content itself. Everpure said this lets IT teams close exposure gaps and reclaim storage capacity before granting AI agents access to shared files.
DeepReduce compression expands usable capacity without hardware upgrades
The company's Always-On DeepReduce technology continuously scans storage blocks across FlashBlade systems for sub-block similarities that standard deduplication tools miss, including in pre-compressed content. Everpure said the feature expands usable capacity automatically without affecting write performance, reducing both hardware footprint and cross-cloud costs.
Reference architecture targets predictable AI spend using open weight models
Everpure introduced a reference architecture built on open weight models designed to give enterprises greater control over data while cutting reliance on external API token usage. Prakash Darji, General Manager of Data & Digital Experience at Everpure, said enterprise AI has stalled not due to model limitations but because data has not been ready for real-time, autonomous agents.
Deployment of the new capabilities is designed to run through Everpure's existing Pure1 console, avoiding separate management servers or extended professional services engagements, according to the company.