# Rerun

A data layer for physical AI that logs, queries and visualizes multimodal robotics data, such as images, point clouds and joint states, in a real-time viewer while streaming the same data into training.

Rerun is a data layer for physical AI: it ingests multi-rate, multimodal data, including images, point clouds, coordinate transforms, time series, joint states and video, from sources such as robot logs, human-data capture rigs, simulation and web video, in formats including MCAP, RRD and LeRobot, storing it on column-chunk storage purpose-built for multi-rate physical data.

Its built-in viewer renders all of this in sync in real time, so a user can scrub through episodes, compare sensors side by side, and watch a computer-vision pipeline run live; the same data can also be queried with dataframes or SQL, and streamed directly into model training without separate export jobs or stale copies. It provides SDKs for Python, Rust and C++, and its viewer can run in a browser or as an installed binary.

It also ships a set of agent skills that help a coding agent write correct Rerun code, and installs through pip, cargo or a C++ integration depending on the language in use.

Website: https://rerun.io/