AI-ready data
Data for your AI applications, without a second store.
Your documents stay in your own Volumes. RunningLake turns them into rows, keeps their embeddings as columns of ordinary Iceberg tables in your bucket, and lets the engines search them in SQL.
AI-ready data PLANNED Reverse ETL PLANNED
How it works
Documents become rows.
A pipeline step reads the documents in a Volume and writes one row per chunk to an ordinary Iceberg table, with a reference back to the file it came from. The files themselves stay where they are.
Embeddings become columns.
Each embedding is a column of numbers in an open table in your bucket, so it is governed and kept like any other column, and your catalog decides who may read it.
Search is SQL.
Similarity search runs in SQL on the engines, in the language your analysts already write.
A vector database is a destination.
When an application needs a vector database such as Pinecone or Milvus, RunningLake keeps it in sync through a registered destination, with your masks applied before anything leaves. Your tables stay the record.
What we leave out
A vector database of our own, and a fourth table format. Every copy of your data that lives somewhere else is one more place to secure, so the record stays in the open tables you already own.