Spiral builds data infrastructure for AI teams. Its central product is an analytical database designed for multimodal, multi-rate data streams, operating on top of the open source Vortex file format. The system is intended to address a specific constraint in modern model training: data loading that cannot keep pace with accelerators. It offers fast queries for exploration and throughput sufficient to saturate GPUs during training.
A second product is a multimodal data platform that converts raw robot data into tested, versioned datasets for training and evaluation, with each iteration reversible and shareable. The company's technical domains span analytical database systems, data formats, time-series data, robot data pipelines, dataset versioning and curation, and data pipeline engineering. Fields of application include weather and climate, financial data, genomics, point clouds, video, images and robotics.
Spiral maintains offices in New York City, San Francisco and London, and is backed by General Catalyst and Amplify Partners. Its work is aimed at frontier AI teams and AI/ML researchers.
