Rerun

Open source alternative to Foxglove

11.8K stars
Apache-2.0
Active this week
Last commit today

Last verified Nov 14, 2026

Visualize and query multimodal data from robotics and physical AI pipelines, with SDKs for Python, Rust, and C++. Self-hosted, Apache-2.0 licensed.

Visit website GitHub11.8K

Read the full breakdownWho it is for, how it works, trade-offs, the quick start and the FAQ

who it's for

Who Rerun is for#

Robotics engineers debugging sensor pipelines

Log every sensor stream during a test run with rr.log() calls, then replay the recording to see exactly what the robot perceived at the moment of failure. Compare RGB camera, depth image, lidar, and segmentation mask side by side at any timestamp.

Skip if:

Your debugging needs are met by text logs or a single-modal time-series database. The setup overhead of adding Rerun SDK calls to your codebase is not worthwhile if you have no need to compare synchronized multimodal streams.

ML researchers building training datasets from recordings

Query a catalog of robot recordings with SQL or dataframe queries to select the training mix, then stream it directly to GPU jobs without intermediate export files. The rrd format preserves the original multi-rate structure so derived columns can be added without reprocessing raw data.

Skip if:

Your training data is already in a structured tabular format or a standard image dataset format. Rerun's value is in extracting and organizing multi-rate sensor data; structured tabular data does not require it.

CV teams running live pipeline instrumentation

Instrument SLAM, hand tracking, motion retargeting, or any CV pipeline with Rerun logging, then watch the intermediate representations in the viewer as the pipeline runs. No separate visualization tool or export step is needed between pipeline stages.

Skip if:

Your pipeline output is a single image or metric per frame with no need to compare intermediate representations across time. A simpler logging approach is sufficient for single-modal, single-rate outputs.

Teams migrating from RViz or MCAP-based workflows

Rerun reads MCAP files directly, so existing ROS bag recordings can be imported without conversion. Unlike RViz, Rerun stores recordings for replay, supports SQL queries, and connects to training pipelines, extending the value of recordings beyond live topic monitoring.

Skip if:

Your team runs a pure ROS environment that only needs live topic monitoring with no replay, query, or training use case. RViz covers that narrower scope with less setup cost.

the problem

The problem it solves#

Robotics and physical AI systems generate data that standard debuggers cannot handle. When a robot misidentifies an obstacle, the failure rarely lives in a stack trace. It lives in the interaction between a camera feed, a lidar scan, a segmentation model, and an odometry estimate, each running at different rates and timestamps. Logging text tells you what the system thought; seeing the data tells you why it failed.

The same recording that caught the bug is also the raw material for the next training dataset. Teams typically manage this by exporting clips, running separate annotation tools, and writing glue scripts to convert between formats. That workflow breaks every time a schema changes or a new sensor is added.

how Rerun solves it

How it solves it#

Time-aware synchronized viewer

The built-in viewer renders all sensor streams in sync and in real time. Scrub back to any frame to compare a camera feed side-by-side with its lidar scan and segmentation output. The same viewer works for live debugging of a running system and offline review of stored rrd recordings.

Full dataframe and SQL queries

Run any SQL or dataframe query across a recording catalog, drilling down to specific columns, time ranges, and values inside the recordings, not just file metadata. Post-processing, annotations, and derived columns can be added as easily as adding columns to a table, without reprocessing raw recordings.

Direct training dataset streaming

Express a training dataset mix as a query and stream it directly to GPU training jobs with no intermediate export step. The dataloader is column-aware and video-codec-aware, so training runs on the rrd recordings as stored.

Multi-rate columnar storage

Data is stored as column-chunks in rrd files, designed for the access patterns of physical AI pipelines where different sensors log at different rates. This keeps every stream time-indexed and queryable without the flattening that general-purpose time-series databases impose.

Broad format support

Rerun ingests its native rrd format, MCAP files (the standard robotics interchange format for ROS bags and more), and LeRobot datasets. The same viewer and query layer works across all supported formats.

SDKs for Python, Rust, and C++

Log data directly from application code using official SDKs for Python, Rust, and C++. The Python SDK bundles the viewer binary so the viewer starts automatically when you call rr.spawn(). Rust and C++ users install the CLI viewer separately via cargo install rerun-cli.

strengths · trade-offs

Strengths and trade-offs#

Strengths

Apache-2.0 and MIT dual-licensed coreThe entire SDK and viewer in the repository are Apache-2.0 and MIT dual-licensed, meaning you can self-host, fork, modify, and use them commercially with no restriction. No features are locked to a paid tier in the open source repo.
Active development with 11,791 GitHub starsThe repository had its most recent push on November 13, 2026, and the team actively contributes upstream to egui and wgpu, the rendering stack Rerun depends on. Active upstream engagement keeps the viewer's graphics layer current with the broader Rust ecosystem.
No export step between recording and trainingUnlike workflows that require exporting clips to separate training pipelines, Rerun streams data directly from recordings to GPU training jobs. The query that selects the dataset mix is the only configuration needed between raw recordings and training.
Rust core for real-time multimodal playbackRerun's core storage and viewer are written in Rust, designed for the throughput and latency demands of real-time multimodal data. The Rust foundation is what makes synchronized playback of high-rate sensor streams practical on a single machine.

Trade-offs

-Viewer performance degrades with many entitiesThe viewer slows down when a recording contains a large number of entities. This is a known limitation tracked in GitHub issue #7115. Scenes with hundreds of tracked objects at high frequency are more likely to hit this ceiling than typical robot debugging sessions with a small set of sensors.
-Multi-million point clouds are slow to renderPoint clouds above a few million points per frame trigger performance issues tracked in GitHub issue #1136. Full-resolution lidar scans from automotive-grade sensors may require downsampling before visualization performs acceptably.
-API is still evolving; breaking changes expectedThe README explicitly states that the API is still evolving and breaking changes should be expected. Teams building on Rerun as a library dependency need to pin versions and review the changelog before upgrading.
-Rerun Hub is not yet publicly availableRerun Hub, the commercial team-scale catalog backend that adds production indexing, shared catalogs, and multi-user data management, is available only to design partners as of the repository README. Teams that need that layer of infrastructure are waiting for a general availability date.

versus alternatives

Rerun vs alternatives#

Rerun vs Foxglove

Rerun and Foxglove both target the robotics data visualization space, but they serve it from different positions. Foxglove is a proprietary commercial product; Rerun is an open source data layer you deploy on your own infrastructure.

Feature	Rerun	Foxglove
License	Apache-2.0 / MIT	Proprietary
Deployment	Self-hosted	Commercial managed product
Scope	Visualization, query, and training integration	Visualization, search, and curation

Rerun is the better fit for teams that need data to stay on their own infrastructure, that want SQL/dataframe queries and direct training dataset streaming in one tool, or that need to instrument pipelines in Python, Rust, or C++. The Apache-2.0 and MIT licenses mean no per-seat costs and no dependency on a third-party service for the core data workflow.

Foxglove is a commercial product designed for visualizing, searching, and curating multimodal sensor data. Teams that prefer a managed, vendor-supported product without infrastructure overhead, and whose primary need is data visualization and curation rather than training integration, should evaluate Foxglove as a commercial alternative to the self-hosted setup Rerun requires.

install · quick start

Quick start#

bash

Install the Python SDK with pip, which includes the viewer binary; for Rust, add the crate with cargo.
```bash
pip install rerun-sdk
cargo add rerun
```

tech stack · detected from GitHub

What it's built on#

Languages: C++PythonRustTypeScript
Frameworks: React
Tooling: esbuild

frequently asked

FAQ#

Is Rerun fully open source?

Yes. The SDK, viewer, and storage format in the GitHub repository are Apache-2.0 and MIT dual-licensed. The only commercial product is Rerun Hub, a team-scale catalog backend currently available to design partners only. The open source SDK has no feature limits and no usage caps.

What data formats does Rerun support?

Rerun supports its own rrd columnar format, MCAP files (the standard robotics interchange format used for ROS bags and other sensor logs), and LeRobot datasets. The rrd format stores multi-rate, multimodal data as column-chunks and is the native format for recording and replay. MCAP files can be imported directly for visualization and query.

How does Rerun compare to RViz?

RViz is a pure visualization tool that shows the latest sensor data from a live ROS topic feed. Rerun is a data layer: it records data to a file, indexes it for time-range queries, and supports replay, dataframe and SQL queries, and streaming to training pipelines. Rerun covers the RViz visualization use case as a subset, but requires logging instrumentation in the application rather than subscribing to ROS topics directly.

Does Rerun work for both real-time and offline data?

Yes. The viewer supports live streaming from a running application via rr.spawn() or rr.connect_grpc(), and offline playback from rrd files. The same recording can be used for live debugging during a test run and for offline analysis, SQL queries, and training dataset extraction afterward.

How stable is the Rerun API?

The project notes in its README that it is under active development and breaking changes should be expected. Teams depending on Rerun as a library should pin to a specific version and review the changelog before upgrading. The repository has 1,214 open issues (pull requests excluded), reflecting ongoing active development.

also worth a look

Similar open-source tools#

Aleph

InData & Analytics

Python CLI for satellite imagery, terrain, buildings and Street View.

867 GitHub stars
Licence MIT

Unlimited-OCR

Alternative toGoogle Document AI

Document OCR with unlimited page depth, MIT licensed, self-hosted.

31,159 GitHub stars
Licence MIT

Marker

Alternative toUnstructured.io

Convert documents to clean markdown and JSON with open source AI.

40,861 GitHub stars
Licence Apache-2.0

Chat2DB

Alternative toDataGrip

AI text-to-SQL client for 40+ databases; local-first, self-hosted

28,364 GitHub stars
Licence Apache-2.0 with additional commercial restrictions (LicenseRef-Chat2DB)

MinerU

Alternative toLlamaParse

Document parsing for LLM workflows, OCR, and agent reading.

83,519 GitHub stars
Licence Apache-2.0 with commercial threshold and attribution terms

Duckle

Alternative toFivetran

Self-hosted ETL on DuckDB with 385 connectors and no per-row billing.

1,563 GitHub stars
Licence Apache-2.0

by category

Open source tools in these categories#

533

12,987 GitHub stars
Licence AGPL-3.0
#Email

rakazo

Alternative toGrok Bot

AI teammates you own: your keys, your model, your machine.

7,869 GitHub stars
Licence Apache-2.0
#AI Agents

SnitchFeed

Live monitoring on LinkedIn, X, Reddit & more. Agentic social listening for startups.

Marketing & Growth

Advertise your product

The projects we add and every new guide. Free, unsubscribe any time.

Find open source alternatives to paid software

Rerun vs Foxglove

Feature	Rerun	Foxglove
License	Apache-2.0 / MIT	Proprietary
Deployment	Self-hosted	Commercial managed product
Scope	Visualization, query, and training integration	Visualization, search, and curation

Who Rerun is for#

Robotics engineers debugging sensor pipelines

Skip if:

ML researchers building training datasets from recordings

Skip if:

CV teams running live pipeline instrumentation

Skip if:

Teams migrating from RViz or MCAP-based workflows

Skip if:

The problem it solves#

How it solves it#

Time-aware synchronized viewer

Full dataframe and SQL queries

Direct training dataset streaming

Multi-rate columnar storage

Broad format support

SDKs for Python, Rust, and C++

Strengths and trade-offs#

Strengths

Trade-offs

Rerun vs alternatives#

Rerun vs Foxglove

Quick start#

What it's built on#

FAQ#

Similar open-source tools#

Open source tools in these categories#

AI & Machine Learning

Developer Tools

Data & Analytics

Find open source alternatives to paid software

Who Rerun is for#

Robotics engineers debugging sensor pipelines

Skip if:

ML researchers building training datasets from recordings

Skip if:

CV teams running live pipeline instrumentation

Skip if:

Teams migrating from RViz or MCAP-based workflows

Skip if:

The problem it solves#

How it solves it#

Time-aware synchronized viewer

Full dataframe and SQL queries

Direct training dataset streaming

Multi-rate columnar storage

Broad format support

SDKs for Python, Rust, and C++

Strengths and trade-offs#

Strengths

Trade-offs

Rerun vs alternatives#

Rerun vs Foxglove

Quick start#

What it's built on#

FAQ#

Similar open-source tools#

Open source tools in these categories#

AI & Machine Learning

Developer Tools

Data & Analytics

Find open source alternatives to paid software