product · arraylake

a data lake for tensors

Arraylake is a managed data lake platform for collaborating on multidimensional arrays and metadata in the cloud. Enterprise catalog and governance, built for humans and agents.

book a demo read the docs

the problem

Data warehouses only understand tables

Existing cloud data warehouses and data lake platforms are built for rows and columns. But most scientific data is best represented in a richer data model — multidimensional arrays with rich metadata. That data model is everywhere in weather, climate, and geospatial work, and throughout bioinformatics, materials science, physics, and engineering.

So teams build their own. Every organization working with this data ends up maintaining a bespoke, in-house stack, and each one pays the same costs:

No catalog — nobody can say what data exists, where it lives, or whether it is current.
Credentials everywhere — cloud storage keys spread across laptops, notebooks, and CI jobs.
No version history — datasets change silently underneath the analyses that depend on them.
DevOps drag — scientists spend their time operating infrastructure instead of doing science.

what arraylake does

A home for your scientific data in the cloud

Arraylake catalogs multidimensional data stored in cloud object storage. Your data lives in Icechunk repositories in your own bucket; Arraylake holds the metadata about those repositories and the record of who may see and change them.

Data catalog

A central catalog of every array asset in your organization, with a native understanding of the Zarr data model. Browse groups, arrays, dimensions, and attributes in the web app — and see how each dataset got there and how it is changing over time.

Access control and governance

Authenticate with single sign-on or your own identity provider. Assign repository permissions by role, issue API keys and service accounts for automation, and audit any state of the data through its immutable commit history.

Credential vending

Arraylake brokers scoped, short-lived cloud credentials on every request, so your users and pipelines never hold long-lived storage keys. Access is granted per repository and revoked centrally.

Bring your own bucket

Configure S3, Google Cloud Storage, Azure, Cloudflare R2, or any S3-compatible endpoint. Your bytes stay in your account under your policies — or let us manage storage for you.

Version control across the lake

Every repository is an Icechunk repo: one shared version history, branches for staging changes, tags for versions that will never move, and time travel to any prior snapshot.

Zero-copy ingestion

Catalog existing NetCDF, HDF5, GRIB, and TIFF archives as virtual datasets. The bytes stay where they are; Arraylake presents them as analysis-ready Zarr.

in practice

Xarray on one side, governance on the other

Arraylake meets you in the tools you already use. Open a repository by name and read it with Xarray; write to it inside a transaction that either lands completely or not at all.

import arraylake as al

client = al.Client()
repo = client.get_repo("my-org/weather-forecast")

# Read the current state of the data — credentials vended by Arraylake,
# no cloud keys in your environment
session = repo.readonly_session("main")
ds = xr.open_zarr(session.store, group="forecast", consolidated=False)

# Write a new forecast cycle as one atomic transaction
with repo.transaction("main", message="2026-08-23 00Z cycle") as store:
    ds_new.to_zarr(store, group="forecast", mode="a", append_dim="time")