Rerun SDK
Rerun SDK is an open-source, code-first database and visualization toolbox for multimodal, temporal data, targeting robotics, computer vision, and AI teams.
Publisher review
Rerun SDK is an open-source, code-first database and visualization toolbox for multimodal, temporal data, targeting robotics, computer vision, and AI teams. It provides a logging SDK in Python, Rust, and C++ that streams structured data—such as Points3D, Images, and Transform3D archetypes—to either a native desktop viewer (Linux, macOS, Windows) or a 32-bit Wasm web viewer. The 0.25 release added transparency, syntax highlighting, and improved table support with filtering; the 0.33.0 release introduced a headless viewer mode for automation and LLM usage, push-down filtering on chunk processing for optimized data extraction, and an enhanced state timeline view with better support for numbers and booleans.
Rerun also offers a catalog server for persistent storage and indexing of large-scale data, organized into datasets and segments, served via the redap protocol, plus a Python catalog SDK for querying and manipulating stored data. The web viewer runs single-threaded and is limited to ~2 GiB memory in practice, making it slower and less capable than the native app. MCAP support is explicitly marked as early/experimental, with limited capabilities compared to competitors.
Rerun competes directly with Foxglove, which offers a full platform with live and recorded data workflows, MCAP as a first-class format, enterprise data governance (SOC 2 Type II, SSO), and team collaboration features like Devices, Events, and Timeline. Rerun’s strengths include its easy-to-use multimodal logging, multiple language SDKs, and both native and web viewers, but it lacks Foxglove’s mature data management, multi-file merged timelines, and enterprise security controls.
How it works
-
Multimodal logging SDK
Logs structured archetypes (Points3D, Image, Transform3D) from Python, Rust, and C++ directly to the Viewer or .rrd files.
-
Native and web viewers
Desktop app for Linux/macOS/Windows and a 32-bit Wasm web viewer; the web viewer is single-threaded with ~2 GiB memory limit.
-
Catalog server with redap protocol
Persistent storage and indexing for large-scale data, organizing recordings into datasets and segments, served via the Rerun Data Protocol.
-
Headless viewer mode
Introduced in 0.33.0, enables automation and LLM usage without a graphical interface, running the viewer in headless mode.
-
Push-down filtering on chunks
Optimizes data extraction by filtering at the chunk processing level, reducing unnecessary data transfer and improving performance.
-
State timeline with numbers/booleans
Enhanced timeline view in 0.33.0 supports numeric and boolean components, improving debugging of state changes over time.
-
Improved table with filtering
0.25 release added table support with filtering capabilities, plus transparency and syntax highlighting for better data inspection.
Strengths and trade-offs
Strengths
- Supports logging and visualization from Python, Rust, and C++ with a consistent archetype-based API, reducing integration effort across teams.
- Provides both a native desktop viewer (full performance) and a web viewer (embedding in Jupyter notebooks or web apps) for flexible deployment.
- Offers a catalog server for persistent storage and indexing of large-scale data, enabling querying and transformation via the Python catalog SDK.
- Introduces headless viewer mode and push-down filtering in version 0.33.0, enabling automated workflows and efficient data extraction for LLM pipelines.
Trade-offs
- Web viewer is limited to 32-bit Wasm with ~2 GiB memory and single-threaded execution, making it significantly slower and less capable than the native viewer.
- MCAP support is explicitly marked as early/experimental with limited capabilities, unlike Foxglove where MCAP is a first-class format for record, ingest, and streaming.
- Lacks enterprise-grade data management features such as SSO, SOC 2 Type II compliance, multi-file merged timelines, and team collaboration tools (Devices, Events) that Foxglove provides.
- No built-in live ROS bridge; users must integrate via SDK logging or MCAP layers, adding complexity for teams already using ROS 1/2.
Pricing context
Open source under an MIT-like license; Rerun Hub (managed catalog server) is a paid offering but no public pricing tiers are listed.
Getting started with Rerun SDK
-
Install Rerun SDK
Install the Rerun SDK via pip for Python: `pip install rerun-sdk`. For Rust, add `rerun` to your Cargo.toml. For C++, download the prebuilt library from the GitHub releases page. Verify installation by importing `rerun` in your code.
-
Log multimodal data
In your Python script, import rerun and call `rr.init('my_app')` to start a recording. Use `rr.log('points', rr.Points3D(positions))` to log 3D points. Stream data to the native viewer by running `rr.spawn()` before logging.
-
Configure viewer mode
Set the viewer mode by calling `rr.spawn(viewer='native')` for the desktop app or `rr.serve()` to launch a web viewer. For headless automation, use `rr.start()` without spawning a viewer, then save data to an .rrd file with `rr.save('output.rrd')`.
-
Inspect data in table
Open the native viewer and select a logged entity. Click the Table tab to view its data in a sortable, filterable table. Use the filter bar to narrow down rows by component values, and enable transparency for overlapping points in the 3D view.
-
Set up catalog server
Deploy the Rerun catalog server by running `rerun catalog serve --data-dir /path/to/data` on your server. Use the Python catalog SDK to connect: `from rerun_catalog import CatalogClient; client = CatalogClient('http://localhost:51234')`. Query datasets and segments via `client.query()`.
Frequently Asked Questions
What is Rerun SDK and what does it do?
Rerun SDK is an open-source, code-first database and visualization toolbox for multimodal, temporal data. It targets robotics, computer vision, and AI teams, offering logging SDKs in Python, Rust, and C++ to stream structured data to native or web viewers.
What programming languages does Rerun SDK support for logging data?
Rerun SDK provides logging SDKs in Python, Rust, and C++. These allow you to log structured archetypes like Points3D, Images, and Transform3D directly to the viewer or to .rrd files, reducing integration effort across different teams.
What are the main differences between Rerun's native and web viewers?
The native viewer runs on Linux, macOS, and Windows with full performance. The web viewer is a 32-bit Wasm build that is single-threaded and limited to about 2 GiB memory, making it slower and less capable for large datasets.
What new features were introduced in Rerun SDK version 0.33.0?
Version 0.33.0 added a headless viewer mode for automation and LLM usage, push-down filtering on chunk processing to optimize data extraction, and an enhanced state timeline view with better support for numeric and boolean components.
How does Rerun SDK compare to Foxglove for data visualization?
Rerun is open-source with easy multimodal logging and multiple SDKs, but lacks Foxglove's enterprise features like SOC 2 Type II, SSO, multi-file merged timelines, and team collaboration tools. Foxglove also treats MCAP as a first-class format, while Rerun's MCAP support is experimental.
What is the Rerun catalog server and how does it work?
The catalog server provides persistent storage and indexing for large-scale data, organizing recordings into datasets and segments. It is served via the redap protocol and includes a Python catalog SDK for querying and manipulating stored data.
Alternatives
How Rerun SDK compares
Direct head-to-head against 2 competitors. Picked by 7wData.
Rerun SDK
- Pricing
- Open source under an MIT-like license; Rerun Hub (managed catalog server) is a paid offering but no public pricing tiers are listed.
- Target
- Rerun SDK is an open-source, code-first database and visualization toolbox for multimodal, temporal data, targeting robotics, computer vision, and AI teams.
- Strength
- Supports logging and visualization from Python, Rust, and C++ with a consistent archetype-based API, reducing integration effort across teams.
- Watch for
- Web viewer is limited to 32-bit Wasm with ~2 GiB memory and single-threaded execution, making it significantly slower and less capable than the native viewer.
Foxglove
- Pricing
- Custom/Contact sales
- Target
- Robotics and Physical AI teams
- Deployment
- Cloud, On-prem
- Strength
- Full platform for live streams, recorded data, and fleet operations
- Watch for
- Complex setup for advanced features
MCAP
- Pricing
- Open source
- Target
- Robotics data logging
- Deployment
- Local, Cloud
- Strength
- Open, serialization-agnostic container for timestamped pub/sub data
- Watch for
- Limited visualization capabilities
User reviews
No user reviews yet. Be the first to write one.
Sources
Reporting on this tool draws on these publicly available sources.