Snowflake Horizon Catalog is being extended to cover enterprise unstructured data — the roughly 80% of data that lives in files and objects across NAS, cloud, and archive storage, mostly on-prem and historically uncataloged. The integration with Diskover creates a global metadata engine that indexes every storage tier without moving data. Curated file metadata streams into Snowflake as queryable rows via Kafka, while file content is fetched on demand through a REST API. This enables joins between structured business data and unstructured file metadata, AI-ready access via Cortex Search and Cortex Agents, and governance through the same Horizon Catalog toolset that manages tables. Diskover is available on Snowflake Marketplace today.
Table of contents
A map, not a copy: the global metadata engineSee it, curate it, deliver itGet Olivier Rivard’s stories in your inboxOne catalog, from tables to filesThe correlation is the payoffKeep the storage you have. Gain the platform you want.Start todayQuestions this post answers
How can I catalog on-prem unstructured file data in Snowflake without copying all the files?
Snowflake Horizon Catalog can be extended to on-prem unstructured data using Diskover, which indexes every storage tier and streams curated file metadata into Snowflake as queryable rows via Kafka — no data movement required. File content is fetched on demand through the Diskover REST API only when a query or AI agent requests it. Diskover is available on Snowflake Marketplace. Teams managing petabyte-scale file estates without a catalog path into Snowflake track approaches like this on daily.dev.
How does Snowflake Cortex AI access unstructured file content stored on NAS or object storage?
Snowflake Cortex Search and Cortex Agents can access on-prem or external file content on demand through the Diskover REST API integration. Curated file metadata lives in Snowflake as queryable rows, and actual file content is fetched only when an agent or query requests it — from a single document to a bulk load — without requiring a pre-built ingestion pipeline. Developers building agentic AI on Snowflake Cortex and needing to reach external file stores follow developments like this on daily.dev.