Toolin.ai
Rerun

Rerun

Officially listed

Multimodal data visualization and logging platform for physical AI, robotics, and spatial computing workloads.

views 72favorites 0Free
Visit website

Rerun

Rerun is a multimodal data infrastructure built by a Swedish team for physical AI and robotics, adopted by top AI companies including Meta, Google, and Hugging Face. It provides a unified data layer spanning data capture, storage, visualization, and model training, solving the core pain point of debugging multi-sensor data in robotics development. In March 2025 it closed a $17 million seed round, and its GitHub has earned 15.6k+ stars.

Core Capabilities

  • Multimodal data logging: Record images, point clouds, tensors, 3D scenes, and other data types through the Python/Rust/C++ SDKs
  • Time-aware visualization: A GPU-accelerated interactive viewer supporting timeline playback and frame-by-frame inspection
  • Efficient data queries: DataFusion DataFrames and SQL interfaces, with columnar storage optimized for query performance
  • Direct training integration: An experimental PyTorch Dataloader can stream data from a Rerun catalog into model training
  • ECS architecture: An Entity-Component System design that flexibly extends to custom data types

Who It's For Robotics algorithm debugging, autonomous driving perception validation, computer vision development, embodied AI training data preparation, and multi-sensor fusion analysis. Especially suited to development workflows that need time playback and offline analysis.

Unique Advantages Compared with RViz, it offers a decoupled architecture and cross-platform support; compared with Foxglove, it is lighter with a strong open-source commitment (Apache-2.0/MIT); compared with TensorBoard, it natively supports 3D and spatial data. Focused on delivering an extreme performance experience for the development phase.

Editor's Review Robotics developers report "super smooth rendering" and "visualizing the KITTI dataset in under half an hour." Written in Rust for guaranteed performance, with an open-source strategy that has earned trust. Note that the API is not yet stable ahead of a 1.0 release, but it is already the most exciting innovation in the physical AI toolchain. Rating: 4.5/5.

Pricing

### 💰 定价模式：开源免费 + 商业订阅 **起步价**：免费（开源版） #### 主要方案 - **开源版（SDK）**：$0 - 完整 Python/Rust/C++ SDK、交互式查看器、SQL查询、PyTorch dataloader、Apache-2.0/MIT 双授权 - **商业版（Rerun Hub）**：定制报价 - 托管 catalog、团队协作、SSO、专属支持、数据存储在客户自己的 S3 桶 #### 试用/其他信息 学术实验室和物理 AI 研究团队可申请优惠。单租户部署在客户选择的云区域。 — Visit website

FAQ

Is Rerun free?

The open-source SDK is completely free (dual-licensed under Apache-2.0/MIT) with full functionality. The commercial Rerun Hub offers hosted catalogs and team collaboration with custom quotes.

What are Rerun's main features?

Core features include multimodal data logging (images/point clouds/tensors), GPU-accelerated visualization, timeline playback, SQL and DataFrame queries, PyTorch training integration, and an extensible ECS architecture.

What is the difference between Rerun and RViz?

Rerun decouples recording from visualization, supports offline playback and cross-platform use, and does not depend on ROS middleware. RViz suits real-time ROS debugging; Rerun suits time-based playback and non-ROS environments.

What is the difference between Rerun and Foxglove?

Rerun is lighter, has no cloud dependency (in its open-source edition), and carries a strong open-source commitment. Foxglove offers enterprise-grade collaboration and fleet management. Rerun fits individual developers and small teams; Foxglove fits large enterprises.

Which programming languages does Rerun support?

Rerun offers SDKs in Python, Rust, and C++, all officially maintained. Install Python via pip, Rust via cargo, and integrate C++ through CMake.

Can Rerun be used in ROS projects?

Yes. Rerun provides ROS 2 integration examples: write a Python bridge node that subscribes to ROS topics and logs them. It supports TF2 transforms, URDF model loading, and MCAP file import (experimental).

Is Rerun suitable for machine learning training?

Yes. Version 0.32 introduced an experimental PyTorch Dataloader that streams data from a Rerun catalog directly to the GPU for training — especially suited to spatial data training for robotics and physical AI.

Is Rerun's API stable?

Rerun is currently pre-1.0 (version 0.33), so APIs may change. The team ships frequently (monthly or biweekly releases); for production use, watch version compatibility.