Rerun

Rerun, from Rerun Technologies AB, is an open-source data layer for physical AI. The SDK logs multi-rate, multimodal data such as images, point clouds, transforms, time series, joint states and video, then visualizes and queries it. SDKs are available for Python, Rust and C++, and a web viewer can be tried in the browser.

What It Is

Rerun is a toolchain for robotics, computer vision and embodied AI teams. It ingests data from robot logs, human-data rigs, simulation and web video, in formats including MCAP, rrd and LeRobot. A built-in viewer renders the streams in sync so users can scrub episodes, compare sensors side by side and watch CV pipelines run live. The same recordings can be queried with dataframes or SQL.

Workflow: From Recording to Training

Data is stored as column-chunks in rrd files, either logged directly or converted from other formats. The site frames the loop as collect, refine, train and deploy. Users add derived columns and run CV transforms, curate datasets with dataframe or SQL queries, and stream dataset mixes to training. Rerun says this works without an export step. Evaluation of rollouts can be traced back to the underlying data.

SDK and Rerun Hub

The SDK includes the viewer, a query library, a visualization framework, a CLI and the file format. It installs with pip install rerun-sdk for Python, cargo add rerun for Rust, and a separate viewer binary for C++ and Rust. Rerun Hub is a commercial backend for catalog, byte-range indexing and retrieval over object storage. The README says it is currently available only to select design partners.

Status and Tradeoffs

The README states the project is in active development with an evolving API and expected breaking changes. Documented shortcomings include slowdowns with too many entities and with multi-million point clouds. The repository also ships agent skills that help coding agents write Rerun code.