Rerun
Open-source SDK and viewer to log, query and visualize multimodal robotics and physical AI data, plus Rerun Hub for team datasets and training
Overview
Rerun is a data layer for physical AI. Its open-source SDKs for Python, C++ and Rust let robotics and perception engineers log images, video, point clouds, 3D transforms, meshes, tensors, text and time series from running code, then scrub through it all on a shared timeline in a native or web viewer. Data is stored in a columnar .rrd format that can be queried as dataframes, and a PyTorch dataloader streams it into model training. A local MCP server lets AI agents drive a running viewer. The commercial Rerun Hub adds a persistent catalog for team datasets, dataset mixes streamed to GPUs and link sharing, deployed single-tenant on your own S3-compatible storage. Stockholm-based Rerun raised a $17M seed led by Point Nine in March 2025.
The Verdict
Who Should Use Rerun?
Best For
- Robotics and autonomy teams debugging perception and control code
- Researchers building robot-learning datasets
- Engineers who prefer logging from code over GUI-first tools
- Teams that want a free, permissively licensed viewer
Not Ideal For
- ROS-native teams that want drop-in bag playback without a bridge
- Simple 2D charts or business dashboards
- Buyers who need public prices for the team product
What's Great
- Free SDK and viewer under Apache 2.0 or MIT
- One timeline for many data types, from video to point clouds
- Quick to get running from Python, C++ or Rust
- Dataframe queries and a PyTorch dataloader for training
- MCP server for agent control of the viewer
Watch Out For
- ROS data needs a bridge or logging node
- MCAP support is still early
- Less built for team review workflows than hosted rivals, without Hub
- Rerun Hub is contact-sales only
Pricing
View all features & details
Key Features
- Multimodal logging: images, video, point clouds, meshes, tensors
- Timeline viewer, native and in the browser
- Dataframe and SQL queries over recordings
- Blueprints for saved layouts
- PyTorch dataloader
- Hub catalog with dataset mixes streamed to GPUs
Use Cases
- Debugging robot perception
- Curating robot-learning datasets
- Reviewing drone and vehicle sensor logs
How It Compares
| Feature | Rerun | Foxglove | RViz |
|---|---|---|---|
| Approach | Code-first SDK and viewer | Web and desktop review platform | ROS desktop visualizer |
| Open source | Yes | See vendor | Yes |
| ROS native | Via bridge | Yes | Yes |
| Data management | Hub (paid) | Yes | No |
Sources: Foxglove comparison (vendor) · TechCrunch (Mar 2025) · Rerun Hub
Is this your tool? Add a badge