01
Visualize everything
See exactly what your robot saw: synchronized cameras, joint states, and 3D in motion. Debug a hard episode, review a dataset, inspect annotation quality, or compare teleop against policy rollouts.
Rerun SDK
The open-source toolchain
for multi-rate, multimodal data
Use the Rerun SDK to log, store, query, view, and train on physical data. Flexible high-performance building blocks for your data loop.
pip install rerun-sdk
rerun10K+
GitHub stars
3
Languages supported
100%
Open source
A single toolchain
for every stage of the loop
02
Query and transform
Analyze across your whole dataset and reshape it however you need. Derive new data (progress scores, action labels, normalized schemas) and add it as new layers to the dataset.
03
Log, store, and convert
Log straight from your code or convert what you already have from MCAP, ROS 2, or Parquet. From there, one format carries the whole loop.
04
Train
Train on the same data you record and query. A dataset mix is just a query, fed straight to your model, with no separate training set format to maintain.
Inside the SDK
Every part of the Rerun SDK is open source and built on the same data
model, so the pieces fit together. That includes APIs and a rerun CLI for spinning up servers and viewers, inspecting recordings,
and converting files.
Logging library
Idiomatic logging APIs in Python, Rust, and C++, built around a semantic data model for Physical AI.
Column-chunk file format
The .rrd format is designed around column-chunks, and is purpose-built to enable flexible and performant processing of multi-rate, multimodal data.
Query and processing APIs
Query recordings as dataframes or with SQL. Slice the columns, time ranges, and frames you need, and reshape data with declarative chunk processing.
Viewer and visualization framework
A fast and extensible viewer and a declarative framework for building views in code. Run on desktop, in a notebook, in the browser, or render headless for agents or automated jobs.
PyTorch dataloader
Train directly on your .rrd files. A column and video-codec aware dataloader that streams dataset mixes to training without needing to re-export.
Catalog server
The open-source catalog that indexes a folder of recordings, so finding and streaming across many files is a query. The local foundation the whole stack runs on.
One open-source stack
The same workflows and APIs as Rerun Hub, powered by the open-source catalog over files on your local disk.
Experiment loop
Teams win by iterating fast on data composition and modeling while scaling data and compute.
Collect
Refine
Train
Deploy
Rerun SDK
One open-source toolchain runs every stage: visualize, query, transform, and train on multi-rate, multimodal data.
Visualization
SQL / dataframe queries
Post-processing scripts
Local training
Open source, local-first
It all runs on the open-source catalog over plain files on your disk, the same APIs as Hub, no infrastructure required.
Open source catalog
RRD
The community loves building with Rerun
The Rerun SDK is open source and widely used across the leading physical AI teams.
LeRobot
Hugging Face's state-of-the-art robotics project uses Rerun as an integrated part of their visualization tools to inspect and debug training runs.
Brush
A 3D reconstruction engine using Gaussian splatting from DeepMind. Written in Rust, portable, flexible, fast — with Rerun visualizing training.
PyCuVSLAM
The official Python wrapper for NVIDIA's cuVSLAM library. GPU-accelerated visual SLAM and camera tracking for real-time localization.
Project Aria
Meta Reality Labs Research's egocentric AI platform. Rerun visualizes sequences in the Aria Dataset Explorer.
Open source stewardship
We invest in the projects we rely on
Rerun is built on wgpu, egui, Apache Arrow, DataFusion, Lance, and more. We're an active contributor to the open source projects we all love and rely on, upstreaming fixes, performance wins, and APIs that benefit the whole ecosystem.
Start in two minutes
pip install rerun-sdk. The viewer, the format, and the toolchain are open source. Use it for research, production, and everything in between.