Vast unveils InsightEngine, a move to support enterprise AI
Vast released InsightEngine to bring real-time retrieval-augmented generation into its data platform as it prepares for growing enterprise adoption of AI.
Vast Data is pushing deeper into data management with a new offering for secure data ingestion, processing and retrieval for AI workloads that gives users up-to-date data while fine tuning their AI training.
InsightEngine with Nvidia is the first application on Vast Data Platform, introduced in August 2023, and provides data storage, management and analytics. InsightEngine uses Nvidia inference microservices (NIM), which can be used to speed up the deployment of AI, embeds the semantic meaning of data and stores it on the Vast DataBase. In turn, this allows for the file, object or table data to be ready for retrieval and inferencing for AI workloads.
This, to me, is the realization of Vast becoming a much bigger player in the data management space.
Matt KimballAnalyst, Moor Insights & Strategy
Vast has been working toward the goal of becoming more than a storage provider, according to Matt Kimball, an analyst at Moor Insights & Strategy.
"This, to me, is the realization of Vast becoming a much bigger player in the data management space," Kimball said.
Vast also launched its Cosmos initiative to bring together researchers, technology partners, service providers and solutions integrators to help users discover the most effective AI use cases. Cosmos will initially include Vast, Nvidia, Cisco, CoreWeave, Supermicro and Deloitte.
Real-time retrieval
Vast is trying to set itself apart from other retrieval-augmented generation (RAG) offerings, which retrieve external data to improve AI responses. Vast said retrieval works with data that is indexed in customer vector databases or data warehouses -- not real-time data.
By utilizing the NIM agents as data is written to InsightEngine, Vast's DataEngine enables real-time vector embedding or graphing relationships of unstructured data. DataEngine is part of Vast's Data Platform and orchestrates and executes events.
Vast's connectors to external data stores such as Salesforce or SAP enable its Vast DataBase to ingest enterprise data in real time, Kimball said.
"What if I could deploy Vast and it would automatically connect to -- and vectorize -- all of my enterprise data that sits inside of my applications?" he asked.
This is the end state of Vast and InsightEngine, Kimball said.
In the past, Vast has focused on providing infrastructure for AI training, according to Steve McDowell, founder and analyst at NAND Research. Now, it's considering enterprise AI workloads, where customers don't train AI models but use pretrained or foundational models with their own data.
"[Vast] is putting in data management capabilities explicitly targeted toward RAG and fine tuning," McDowell said.
Different data management
Vast uses a disaggregated shared-everything (DASE) architecture, a framework for breaking entire systems down to smaller parts, enabling the company to store trillions of embeddings while supporting real-time data ingestion and real-time similarity search across vector spaces and knowledge graphs for structured and unstructured data in a unified namespace, according to the vendor.
InsightEngine uses a unified data architecture that collapses storage, processing and retrieval together, eliminating the need for external data lakes or SaaS platforms, according to Vast.
Vast is doing a lot to stand out in a crowd of AI offerings, Kimball said, including being the first vendor to integrate NIMs into data embedding. It can also present structured and unstructured data to AI agents from a single source and perform memory-speed searches across the trillions of embeddings for real-time results.
"As enterprises become driven by [AI] agents, the number of embeddings required to support vectors is going to grow exponentially," he said.
Stepping away from just storage, with caveats
Vast's DASE architecture has made the vendor hard to categorize, according to Kimball. Now with its push toward data management in the last couple of years, that task has become even harder, he said. Other vendors such as NetApp are moving in a similar direction.
"It's an identity crisis, of sorts, that the market has to figure out," he said.
From the start, Vast was pointing out inefficiencies in moving data in and out of the storage layer, McDowell said. Vast designed a product that allows customers to do data management and manipulation inside the storage to make everything run faster.
"They already set themselves up as, 'We're not a storage platform, we're a data platform, which means I'm not just storing bits on a box, I'm helping you do something with them,'" he said.
But McDowell did express a word of caution.
"It is also a lock-in," he said. "If I start using the database capabilities, it becomes very sticky and hard to migrate away from."
Adam Armstrong is a TechTarget Editorial news writer covering file and block storage hardware and private clouds. He previously worked at StorageReview.
