AI Search Unlocks Video Archives
VAST Data and TwelveLabs have partnered to bring AI-powered search and analysis to petabyte-scale video archives. The collaboration focuses on semantic search, automated tagging, and event detection for large, secure repositories in government and enterprise. This signals a growing demand from newsrooms for AI tools that can surface value from back-catalogs, not just process new content.
The partnership provides a customer-managed deployment option, moving TwelveLabs' video foundation models onto a customer's own infrastructure managed by the VAST AI Operating System. This move is designed for organizations with large video archives in controlled environments where public cloud solutions conflict with data sovereignty, security, or regulatory requirements. At the core of the collaboration are TwelveLabs' multimodal AI models, Marengo and Pegasus. Unlike traditional systems that rely on text metadata, these models understand video by simultaneously processing visuals, audio, and text, allowing for natural language searches like "find the moment the CEO discusses Q4 earnings" across petabytes of footage. VAST Data provides the underlying AI-native infrastructure, which unifies data storage, a database, and a compute engine into a single operating system. Its platform creates a "global namespace" to access all video files, whether on-prem or in the cloud, and includes a built-in vector database capable of handling the trillion-vector scale needed for these AI models. This "bring the AI to the data" approach avoids the massive cost and complexity of moving exabytes of sensitive video to the public cloud for processing. It specifically targets workflows in media, government, and financial services where video is a long-lived, highly-governed asset that has been historically difficult to search and monetize.
