Spiral

Spiral provides a storage and query engine for compressed multimodal data such as video, audio, images, traces and tabular data. It queries data in place from MCAP, LeRobot, MP4 and Parquet files stored in the customer's own bucket (S3, GCS, R2, MinIO and others), and streams selections directly to GPU training loops. The site describes versioned datasets with time travel to earlier commits, tags or branches.

The site presents the product for teams training robot and physical AI models, and cites examples of such teams in its case snippets. Access is by request: the site says it onboards a few teams at a time. The website does not state pricing or the revenue model; the business model shown is a request-access B2B product.

The company says it was created by the team behind Vortex, an open-source storage format that the site says is a Linux Foundation project. Open job listings are for software engineers (systems) in San Francisco, New York City and London, and a developer relations engineer in New York City, London and San Francisco. The registry lists the company as based in the US. The founding year is not stated on the website.

Product: A storage and query engine for compressed multimodal data that runs queries in place on customer buckets and streams training batches to GPUs.