The data primitives to build, understand, and improve your data loop. Designed for multi-rate, multimodal data, from the first recording to massive scale.
10K+
GitHub stars
3 supported languages
End-to-end
Collection to training
End-to-end models learn from multi-rate, multimodal sequences. So every stage of the loop works on that same data: collect, refine, train, deploy. Rerun is one layer underneath all of it.
Teams win by iterating fast on data composition and modeling while scaling data and compute.
Teleop, fleet logs, human-data rigs, sim, and web video, all in one format.
Add columns and run CV transforms. Curate with dataframes or SQL.
Stream dataset mixes straight to GPUs. No export jobs, no idle waiting.
Evaluate rollouts and trace failures back to the data that caused them.
Open source SDK that makes it easy to build the applications and compute you need to create physical intelligence.
df.select(...).where(...)
for batch in dataloader:
Data catalog and backend for large-scale storage, access, and streaming from object storage. Makes Rerun SDK scale.
The Rerun SDK is the open-source toolchain physical AI teams build on. Rerun Hub is the production backend that connects it to your data at scale.
Open source
Open-source SDK for logging, storing, querying, visualizing, and training on multi-rate, multimodal data. Includes viewer, query library, visualization framework, CLI, and the file format the data layer is built on.
pip install rerun-sdk
rerunCommercial
The production backend for the Rerun data layer. Catalog, byte-range indexing, and retrieval that turns your object stores into a queryable, streamable foundation. Run transforms on the edge or close to the data.
You log, query, visualize, transform, and train with the same open-source SDK and the same APIs either way. Hub doesn't change what you do. It changes what runs underneath.
Build, understand, and improve your data loop on one set of primitives, from your first recording to massive scale. Start with the SDK, or talk to us about Hub.