Rerun
Open-source data layer for Physical AI: log, query, transform, visualize, and train multimodal robotics data on one toolchain.
For robotics work that spans visualization and training, Rerun is the most complete open-source answer available — the SDK is genuinely free under Apache-2.0/MIT, and SQL queries into recording values plus a codec-aware PyTorch dataloader go well past what Foxglove offers. The real decision is Hub: it's contract-priced per deployment after a call, so start on the SDK, measure what file-on-disk workflows cost you in idle GPUs and export scripts, and only then talk numbers.
Verified 39m ago · liveness 81/100 · cite: rightaichoice.com/tools/rerun
- Robotics and embodied-AI teams running one pipeline from raw sensor logs to model training
- Researchers debugging policy rollouts where trajectory, sensor frames, and loss curves must be read together
- Teams scaling from laptop experiments to cloud training without rewriting their logging code
- Academic labs and research groups working on robot learning or 3D reconstruction
- General-purpose BI or business dashboarding — this is built around spatial and temporal sensor data
- ROS-only stacks with no training or dataset side to connect
- Teams that want a fully managed, zero-setup platform from day one with no SDK-first phase
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Rerun if you only need to debug live ROS topics and never train on the data — Foxglove gets you there with less setup.
Hub is priced per-deployment on team size and data scale, so there's no published number to budget against — expect a sales process rather than a checkout page.
The SDK is $0 forever (Apache-2.0/MIT), which puts it well below Foxglove's paid tiers and W&B's per-seat pricing for the logging and visualization piece. Hub is quoted per deployment, so it competes on total data footprint rather than seats — cheap for a small lab, a real budget line for a multi-region production team.
In short
Rerun — Open-source data layer for Physical AI: log, query, transform, visualize, and train multimodal robotics data on one toolchain. Best for Robotics and embodied-AI teams running one pipeline from raw sensor logs to model training, Researchers debugging policy rollouts where trajectory, sensor frames, and loss curves must be read together, Teams scaling from laptop experiments to cloud training without rewriting their logging code. Free to use.
What's new in Rerun
Checked todayAcross the latest 1 update: 1 changelog entry.
What people actually say about Rerun — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
45 mentions across 4 sources (Hacker News, App Store, GitHub, Lemmy) · researched Jul 3, 2026.
Average across the 4 sources that answered — each source counts once, not each post.
- +Unified data pipeline from logging to training in one tool.
- +Open-source SDK with permissive Apache-2.0/MIT license.
- +Efficient columnar storage for high-dimensional time-series data.
- +Interactive 2D/3D viewer ideal for robot sensor data.
- +Integrations with LeRobot and cuVSLAM add real-world credibility.
- −Over 1300 open GitHub issues signal reliability concerns.
- −Steep learning curve for beginners and non-robotics users.
- −Limited community discussion outside GitHub and niche forums.
- −Hub pricing not transparent; potential for unexpected costs.
- −New file format .rrd may not integrate with existing tools.
- • Hub pricing is not public—enterprises may face surprise costs.
- • Self-hosting Hub likely requires significant infrastructure.
Viability Score
How well maintained and how widely used is Rerun? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Log multi-rate, multimodal data from Python, Rust, and C++ SDKs
- Interactive viewer on desktop and in the browser
- 2D, 3D, map, graph, tensor, text, and time-series views in one data model
- Run SQL or dataframe queries into recording columns, time ranges, and values
- Add derived columns and evolve schemas without breaking history
- Store recordings as column-chunks in the columnar .rrd file format
- Convert data in from other formats, including HDF5 and MCAP
- Stream dataset mixes to GPUs with a column-aware PyTorch dataloader
- Declarative blueprint layers for programmatic visualization
- Extend the viewer with your own custom views and tools
- Measurements archetype for scalar series (0.38.1)
- Load local .rrd files through the Viewer catalog (0.38.1)
- Control the viewer time cursor from Python (0.38.1)
About Rerun
Rerun is the open-source data layer for Physical AI, aimed at robotics and embodied-AI teams that need one pipeline from raw sensor logs to model training. Its designers built it for multi-rate, multimodal data — camera frames, LiDAR sweeps, joint states, IMU traces and text all land in a single data model rather than a pile of bespoke scripts. The SDK installs with pip install rerun-sdk and ships Python, Rust, and C++ APIs, so a research notebook and a C++ onboard stack can write to the same recordings. Once data is logged, you work on it in place. Full SQL or dataframe queries reach down into the columns, time ranges, and values inside a recording, not just its metadata. Derived columns and schema evolution let you add annotations or post-processing without rewriting history. Recordings are stored as column-chunks in the .rrd file format, and HDF5 and MCAP paths cover data you didn't log with Rerun itself. A column-aware, video-codec-aware PyTorch dataloader expresses a dataset mix as a query and streams it to GPUs, so training doesn't wait on an export step. Visualization spans an interactive desktop and web viewer rendering 2D, 3D, map, graph, tensor, text, and time-series views, with declarative blueprint layers you define in code and extend with your own views. The 0.38.1 release added a Measurements archetype for scalar series, local .rrd loading through the Viewer catalog, and Python control of the viewer time cursor. Rerun Hub is the commercial side: the same APIs backed by a persistent managed catalog, byte-range indexing across your own S3-compatible object storage, streamed dataset mixes, link sharing, auth and SSO, and single-tenant isolation in a region you pick. The SDK stays free forever under Apache-2.0/MIT. Teams at NVIDIA (PyCuVSLAM), Meta Reality Labs (Project Aria), and DeepMind (Brush) use it, as does Hugging Face's LeRobot.
Behind the Verdict
The pitch that lands is the feedback loop. When your policy rollout, training loss, camera feed, and 6-DoF trajectory live in one viewer, debugging a calibration mismatch stops being a scavenger hunt across three tools. I'd reach for Rerun specifically when the failure you're chasing is temporal and spatial at once — a bad trajectory segment that only makes sense next to the sensor frame that produced it. Where it beats the usual suspects: Foxglove renders sensor data well but has no answer for dataset mixes streaming to your GPUs. Weights & Biases tracks runs but won't draw a point cloud. Peter Mitrano's summary in the community thread is the honest one — plotting loss in Rerun is roughly as easy as W&B, and the advantage shows up when you need an evaluation visualization no standard tool handles. The caveat that matters commercially: Hub is quoted per deployment, so you can't put a number in a budget spreadsheet from the site. That's normal for infrastructure sold to robotics orgs, but it means the SDK is where you should plan to live for a while. Academic labs get discounts — worth asking before you assume the commercial route is priced out of reach. Skip it if you need dashboards for business metrics, or if your stack is ROS-only with no training side to connect. And this is a developer tool: teams wanting a zero-setup managed platform from week one will find the SDK-first posture a climb. Practically, the .rrd format and columnar storage are the load-bearing choices. Schema evolution that preserves history means last quarter's recordings stay queryable after your log schema changes, which sounds boring until you're six months into a project and need to compare two generations of data. That's the reason teams keep it around after the novelty of the 3D viewer
Researching Rerun? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Rerun actually fits — and what changes day-one when you adopt it.
You log camera, LiDAR, IMU, and joint-state streams from a dataset run with the Python SDK, open the recording in the viewer to check time alignment, then query a specific time window with dataframe queries to find the frames where perception drifted.
Outcome: You trace the failure back to the exact recording and frames, then add a derived column with the corrected labels and keep both versions without breaking history.
Your recordings live as .rrd files or in object storage. You express the training mix as a query, and the PyTorch dataloader streams it straight to the GPU — column-aware and video-codec-aware, no export step.
Outcome: Training starts without a separate dataset conversion pipeline, and the same query is the reproducible definition of what the model trained on.
Your team outgrows files on disk. You move to Rerun Hub in your chosen region, with your data staying in your own S3-compatible buckets, and use link sharing so reviewers open the same recordings.
Outcome: Everyone explores and annotates the same recordings, and failures get traced back to source data without ad-hoc file handoffs.
Use Cases
- Visualize multimodal robot logs with synchronized 2D, 3D, and scalar time-series views
- Query historical recordings with SQL or dataframe to debug perception failures
- Stream training data mixes directly to GPUs from object storage without export steps
- Transform raw sensor data with derived columns and share refined datasets across the team
- Convert ARKitScenes scans into recordings that the viewer, dataframe API, and dataloader read directly
- Debug training runs and visualize policy rollouts with the same toolchain
Limitations
- The open-source SDK (Apache-2.0 / MIT) is free forever, but the commercial Rerun Hub is priced per-deployment on a custom contract with no public tiers — you must book a meeting with sales to get a number, which makes budgeting hard before committing.
- Hub is run for you, single-tenant, in the cloud region you choose, with data staying in your own S3-compatible object storage; deploying the data plane into your own account is only available for large teams with specific requirements.
- Some data-format support (e.g.
- MCAP) has been experimental, so long-term archiving of recordings is a consideration.
as of 2026-09-14
Verification history
We have re-verified Rerun 7 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 7 verification passes.
Free to cite with attribution — this page re-verifies continuously.
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Plans compared
For each published Rerun tier: who it actually fits, and what it adds vs. the previous tier. Cross-reference the cost calculator above for projected annual outlay.
Open source SDK
$0
Ideal for
Robotics researchers, academic labs, and engineers who can run the toolchain on their own machine or cluster and store .rrd files on disk.
What this tier adds
Free entry point: the full log, query, transform, visualize, and train toolchain under Apache-2.0/MIT, with a local catalog and community support.
Commercial Hub
Custom
Ideal for
Physical AI teams that have outgrown files on disk and need a managed catalog with shared access and streaming from object storage at production scale.
What this tier adds
Adds everything the SDK leaves you to run yourself: a persistent managed catalog, byte-range indexing over your object stores, dataset mixes streamed to GPUs, link sharing, auth and SSO, and single-tenant isolation in your chosen region.
Where the pricing makes sense
The company stage and team size where Rerun's pricing actually pencils out — and where peers do it cheaper.
The SDK is $0 forever (Apache-2.0/MIT), which puts it well below Foxglove's paid tiers and W&B's per-seat pricing for the logging and visualization piece. Hub is quoted per deployment, so it competes on total data footprint rather than seats — cheap for a small lab, a real budget line for a multi-region production team.
Setup time & first value
How long it actually takes to get something useful out of Rerun — broken out by persona, not the marketing-page minute.
Researchers: minutes to first value — `pip install rerun-sdk`, log a few entities, and the viewer opens. C++/Rust teams: an afternoon to wire the SDK into an existing build and land first recordings. Teams adopting Hub: expect a sales conversation plus a deployment window before the managed catalog is live in your region.
Switching to or from Rerun
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From ROS bag / ROS 2: use the ROS 2 integration and MCAP support to bring existing log topics into recordings.
- →From HDF5 archives: use the HDF5 importer (0.35+) to convert stored datasets into .rrd recordings.
- →From LeRobot datasets: Rerun documents working with LeRobot datasets and can export recordings back to that format.
- →From MCAP files: import MCAP directly using the documented decoders and message formats.
- ↗To LeRobot datasets: Rerun documents exporting recordings to LeRobot datasets when you need that ecosystem's format.
- ↗To a Pareto: for pure ROS visual debugging, Foxglove remains the simpler drop-in for live topic inspection.
Integrations
Resources & Guides
- Documentationrerun.io
Docs · Rerun
Full product docs from rerun.io
- Quickstartrerun.io
Getting Started · Rerun
Get up and running fast from rerun.io
- Conceptsrerun.io
Concepts · Rerun
Core ideas explained from rerun.io
- Documentationrerun.io
Howto · Rerun
Full product docs from rerun.io
- API Referencererun.io
Reference · Rerun
Methods, params, types from rerun.io
- Examplesrerun.io
Examples · Rerun
Working sample projects from rerun.io
- Documentationrerun.io
Changelog · Rerun
Full product docs from rerun.io
- Resourcererun.io
Viewer · Rerun
Helpful link from rerun.io
Tutorials & Learning
YouTube returned 6 videos for “Rerun”, and we withheld 6: 6 could not be judged, because “Rerun” is a single word that other videos use for other things. We are showing none, because we could not prove any of them are about Rerun.
Official links
Tools that pair well with Rerun
Common stack mates teams adopt alongside Rerun, with the specific reason each pairing earns its keep.
OpenAgents
OpenAgents is an Apache-2.0 platform for language agents that analyze data, call 200+ plugins and browse the web.
Ambient.ai
Ambient.ai turns physical security from reaction into prevention with an always-on AI reasoning layer over your existing camera and access-control systems.
Quadratic
Quadratic is an AI spreadsheet where the grid runs Python, SQL, JavaScript, and formulas against live data sources.
Featured Head-to-Head Comparisons
Rerun vs Spider Cloud
Spider Cloud and Rerun serve completely different domains: Spider Cloud excels at fast, cost-effective web crawling for AI agents and RAG pipelines, while Rerun is purpose-built for robotics teams logging and training on multimodal sensor data. Choose Spider Cloud if your need is real-time web data extraction; choose Rerun if you're building Physical AI systems.
Rerun vs Temporal Ai
Temporal AI is the choice for teams building reliable AI agents and complex business workflows that need guaranteed execution and recovery. Rerun excels for robotics engineers needing to log, visualize, and train on multimodal sensor data. Choose Temporal if your pain is crash recovery and orchestration; choose Rerun if your pain is debugging and scaling physical AI data.
Rerun vs Screenplayiq
These are not competitors, and no buyer should be choosing between them. ScreenplayIQ is a paid per-analysis tool for feature-film scripts — it sells box-office prediction, beat sheets, character-arc charts, and PDF coverage reports to writers, producers, and studio execs. Rerun is a freemium open-source data layer for Physical AI: robotics teams log sensor data via Python/C++/Rust SDKs, query it with SQL, and stream .rrd datasets straight to PyTorch. Different budget lines, different departments, different problems. If you're buying script feedback, only one of these is even a candidate; if you're buying a robotics data pipeline, the same is true in reverse.
Alternatives to Rerun
View all
OpenAgents
OpenAgents is an Apache-2.0 platform for language agents that analyze data, call 200+ plugins and browse the web.
Ambient.ai
Ambient.ai turns physical security from reaction into prevention with an always-on AI reasoning layer over your existing camera and access-control systems.
Frequently Asked Questions
Best-of guides
Used Rerun? Help shape our editorial sentiment research.