Rerun Hub. The backend for scaling physical AI.

Move to production without letting local files become the limit. Rerun Hub keeps multimodal data queryable, shared, and ready to scale.

Book a meeting Read docs

Scaling physical AI is complex. Rerun Hub solves it for you.

Exercitation pariatur reprehenderit esse lorem, tempor, nostrud cillum labore excepteur, culpa, et, adipiscing.

A rising dotted curve marks growth from 10 terabytes to 10 petabytes.

Scale to petabytes while keeping recordings fast to browse, query, and stream.

Three teammates discuss the best way to ingest data into Rerun.

Work with a dedicated Rerun team to get production data infrastructure running quickly.

Scale indexing and queries automatically as your collection grows.

Bring the whole team together around the same recordings and derived data.

End-to-end infra for scaling and shipping physical AI

Query

When moving to production managing local files quickly become segment ationis. Wasmus grpcus multimodalus parquetus schemae temporalis seg menta tionis coordinatum.

Query everything with SQL across recordings, sensor values, timelines, and metadata.

A search query narrows a grid of robot camera frames.

Find the moments that matter across thousands of runs, from failures to valuable edge cases.

Three synchronized point-cloud views show the same physical scene.

Get answers without moving data using selective reads that stay fast as your catalog grows.

Transform

When moving to production managing local files quickly become segment ationis. Wasmus grpcus multimodalus parquetus schemae temporalis seg menta tionis coordinatum.

The Rust and WebAssembly logos joined by a plus sign.

Unify every robotics format into one semantic data model built for multi-rate, multimodal data.

Three teammates discuss how to turn raw logs into useful data.

Turn raw logs into useful data with native tools for normalization, alignment, and feature generation.

Three synchronized point-cloud views show derived data from the same scene.

Build on your data without copies by adding annotations and derived signals alongside the original recordings.

Train

When moving to production managing local files quickly become segment ationis. Wasmus grpcus multimodalus parquetus schemae temporalis seg menta tionis coordinatum.

Turn queries into training sets and iterate on dataset composition as quickly as you iterate on models.

Three teammates discuss preparing data for model training.

Feed PyTorch-ready batches directly with native support for numeric data, images, and compressed video.

Robot-arm camera frames stream into an object storage bucket.

Keep your GPUs fed at scale by streaming straight from object storage without an export pipeline.

When moving to production managing local files quickly become segment ationis. Wasmus grpcus multimodalus parquetus schemae temporalis seg menta tionis coordinatum.

Give every team one source of truth for recordings, schemas, annotations, and derived data.

Synchronized robot-arm views include a review annotation.

Make physical data easy to understand with synchronized video, 3D, signals, and annotations.

Two collaborator cursors are labeled José and Guillaume.

Collaborate securely from anywhere while your data remains inside your own object storage.

All your use cases. Supported out of the box.

Collect

Log multimodal data from your sensors, cameras, and simulations into one unified, browsable timeline.

Refine

Filter, transform, and re-label your logged data streams to iterate on model inputs without re-running.

Train

Track losses, metrics, and visual outputs in real time alongside the data that produced them.

Deploy

Validate model outputs against logged ground truth so regressions never reach users.

Production ready

All the certifications you expect for production, with SOC 2 and ISO 27001 to support your security and compliance requirements.

Tenant isolation keeps your data separate and secure, with access controls that align with your organization's security policies.

Reliable availability keeps recordings ready to query and stream, with fast access to your object storage as your workloads grow.

Start with open source. Add Rerun Hub for scaling.

	Open source Free and open forever	Rerun Hub Commercial
Scalability	Limited scalability using local files or custom solutions.	Instantly scale to petabytes using Rerun Hub's performant cloud infrastructure. Automatic scaling.
Data storage	Local `.rrd` files on one machine with an in-process catalog for local work.	Persistent, managed catalog with byte-range indexing across your own S3-compatible object stores. Data stays in your buckets.
Performance	Performant Viewer and SDK, but the back end is limited by local hardware constraints and hand-rolled infrastructure.	Added on top of the SDK and Viewer, cloud-based indexing ensures that queries and lookups remain fast across petabyte-sized datasets.
Collaboration	Single-user; data lives on your local machine with no built-in sharing mechanism.	Team-wide access to the same recordings via authenticated share links. Everyone works from the same data.
Access control	No authentication or access control; access depends on who has the machine.	SSO authentication, authenticated share links, and single-tenant isolation deployed in the cloud region you choose.
Compliance	Audited SDK and Viewer, but the rest is up to you.	End-to-end compliance built into the entire stack. SOC 2 and ISO 27001 certifications, 24/7 monitoring, and continuous audits.
Support	Community support via Discord and GitHub. Open source under Apache-2.0 / MIT.	Commercial support from a shared Slack channel to a dedicated forward-deployed engineer. Academic and research discounts available.

Let's scale

Pinholus webviewus arrowus transformae gridus schemae rrdum. Parquetus originus containerae quaternionae, webviewus sdkum, pathus componentum. Framus containerae pathus.