Rerun

Rerun is an open-source software development kit designed to serve as the data layer for physical artificial intelligence systems, particularly those dealing with multimodal, multi-rate data streams such as robotics, autonomous vehicles, and computer vision applications. It provides a single toolchain that covers the entire lifecycle of data handling: from initial logging and recording, through transformation and querying, to visualization and training. The core functionality revolves around a flexible framework that allows developers to log arbitrary data types, including images, point clouds, 3D meshes, time series, text, and custom geometry, all with precise timestamps and support for multiple data rates simultaneously. One of the key features is its ability to visualize complex datasets in real time or from recorded files. Users can review entire datasets, drill down into detail-level issues, and extend the viewer with custom views and tools tailored to specific pipeline stages. The visualization interface supports interactive 3D scenes, 2D plots, histograms, and more, enabling engineers to inspect sensor data, simulation outputs, and model predictions side by side. Beyond visualization, Rerun offers powerful query and transformation capabilities. It supports full dataframe-like operations and SQL queries over any stored robotics data, allowing users to filter, aggregate, join, and post-process data programmatically. Extensions for annotation, post-processing, and custom pipelines can be added using Python or Rust, integrated with the existing logging primitives. This makes it possible to build annotation tools, compute metrics, or generate training datasets directly from the recorded data. The SDK is available for C++, Python, and Rust, with a simple API that blend