Overview

Rerun is a data layer and visualizer for physical AI and robotics work. It ingests multi-rate, multimodal data such as images, point clouds, transforms, time series, joint states, and video from robot logs, simulation, human-data rigs, and formats like MCAP, rrd, and LeRobot. The built-in viewer renders these streams in sync, so you can scrub episodes, compare sensors, and watch computer-vision pipelines run live. The same recorded data can be queried through dataframes or SQL and streamed to training. SDKs are available for Python, Rust, and C++, with the core written in Rust What it does and who it is for.

Typical users are robotics, simulation, and computer-vision engineers who need a temporal debugger rather than a plain log viewer. For example, when a robot behaves incorrectly, Rerun can show the camera feed, depth images, lidar scan, segmentation output, 3D map, and detections together along a shared timeline example use case. It is not limited to robots: any project with sensors or 2D/3D state changing over time can use it.

Getting started

Python is the shortest path because the Python SDK bundles the viewer:

1. Install the SDK with pip install rerun-sdk. Conda users can use the rerun-sdk feedstock; Rust users can run cargo add rerun.
2. Log a simple recording:

import rerun as rr

rr.init("rerun_example_app")
rr.spawn()  # starts a viewer and connects to it
rr.set_time("frame", sequence=42)
rr.log("path/to/points", rr.Points3D(positions, colors=colors))

3. Inspect the logged points in the spawned viewer. Use rr.save("recording.rrd") to write to disk or rr.connect_grpc() to attach to a remote viewer short example.

C++ and Rust do not bundle the viewer, so they need a separate rerun binary. It can be installed with pip install rerun-sdk or cargo install rerun-cli --locked --features nasm; the nasm feature requires the nasm CLI on your PATH, but it can be skipped at the cost of slower video decoding viewer installation. After installation, rerun --help should work. Start with the Python, C++, or Rust getting-started guides; the browser viewer can open recordings without a local install.

Constraints

Rerun is under active development and the API is still evolving, so breaking changes should be expected. The documentation lists known limits: the viewer slows down with too many entities, and multi-million-point point clouds can be slow status and shortcomings. In long-running or memory-constrained applications, the viewer's in-memory buffer can be resized, including when replacing RViz limit RAM. MCAP ingestion and LeRobot-style data are supported, but users working with other third-party formats should check the ingestion documentation first.

The core is open source under MIT and Apache-2.0. Rerun Hub, a scalable catalog for robotic data, is a separate commercial offering available to select design partners; the repository itself remains free and open source business model.

HotRepo's take

Rerun appears on the Rust daily GitHub list on 2026-11-06 at #12 with 14 stars gained that day. That is a single snapshot, so it does not establish a trend, but it shows the project is still visible in the Rust ecosystem alongside its 11.7k stars. The project also has several Hacker News posts tracking releases from 0.25 through 0.30, but the provided discussions carry only a handful of points and comments. One comment on the 0.30 release notes that arbitrary scalar plotting now covers floats, ints, uints, bools, and nested Arrow data without predefined semantics, and that custom visualizers can be registered into existing 2D/3D/Map views with custom shaders 0.30 discussion. Those are release-specific claims, not independent evaluations of quality.