Rerun

Rerun builds a unified data layer for physical AI, supporting robotics, spatial computing, and embodied AI teams. It enables ingestion of multi-rate, multimodal data and provides tools to log, query, visualize, and train across the data pipeline. The open source Rerun SDK offers logging, storage, querying, transformation, visualization, and training capabilities that developers can run locally and extend. Rerun Hub adds a commercial data catalog and storage engine for larger collections backed by object storage.

Rerun provides an open source SDK that developers can run locally and extend. It offers logging, storage, querying, transformation, visualization, and training tools built on a single data model. Hub is the production backend that connects the SDK to data at scale, providing catalog and storage capabilities. Both use the same data model and APIs, so workflows can grow without replacing tools used during early development.

The tool aims to help teams collect, refine, train, and deploy data and models. It supports end-to-end learning where end-to-end models learn from multi-rate, multimodal sequences, so every stage of the loop works on the same data.

Rerun emphasizes collaboration by enabling teams to work from the same data: visualizations, annotations, and tracing of failures can be shared across a team. The platform is designed to log and transform data, query it efficiently, and train on recordings, with a workflow that scales from local development to larger-scale deployments.

Use cases

• Visualize and debug data across a robotics experiment
• Query recordings with SQL or dataframe operations
• Train models directly on logged recordings without export steps