Rerun is an open-source SDK designed for logging, storing, querying, and visualizing multimodal and multi-rate data, serving as a data layer for physical AI applications.
Source: README View on GitHub →Rerun is gaining attention due to its ability to handle complex multimodal data, providing a comprehensive solution for logging, querying, and visualizing data from various sources like robot logs, simulations, and computer vision pipelines. Its unique technical choice of using columnar storage for multi-rate physical data stands out, offering efficient data management and processing.
Source: Synthesis of README and project traitsRerun can ingest a wide range of data types including images, point clouds, transforms, time series, joint states, and video from various sources and formats, making it versatile for different types of physical AI applications.
Source: READMEThe built-in viewer allows for real-time visualization of data, enabling users to scrub episodes, compare sensors side-by-side, and watch CV pipelines run live, enhancing debugging and analysis capabilities.
Source: READMEData is queryable using dataframes or SQL, and can be streamed directly into training, eliminating the need for export jobs and ensuring data freshness.
Source: READMEThe architecture of Rerun is modular, with separate crates for building, storing, and viewing data. It uses a columnar storage system optimized for multi-rate physical data, and provides SDKs in Python, Rust, and C++ for different application needs. The project utilizes Cargo for dependency management and Rust's edition 2024 for modern language features.
Source: Code tree + dependency filesCenter: project; inner ring: core feature modules; outer ring: key dependencies. Auto-generated from core_features and tech_stack.key_deps.
rerun_pyrerun-clirerun_cRerun is suitable for robotics, simulation, computer vision, and any application involving sensors or signals that evolve over time. It is useful for debugging robots, managing and querying training data, visually debugging live streams or recordings, and creating datasets for training and evaluation.
Source: README0.32.0 (2026-05-13): Chunk Processing, Pytorch dataloader, Dataset Review
Source: GitHub ReleasesRerun is a promising project for teams and individuals working on physical AI applications that require robust data logging, querying, and visualization capabilities. Its focus on multimodal data and efficient data management makes it a valuable tool for complex AI systems.