PolyStore

Framework-agnostic storage primitives for scientific applications.

PolyStore owns backend interfaces, execution-local backend registries, FileManager routing, storage formats, ROI values, virtual-workspace source references, and viewer-streaming payload mechanics. Applications retain ownership of domain artifact names and materialization policy.

Quick start

Pass an explicit mapping of backend names to instances. FileManager has no global fallback and does not infer a backend from a path.

from pathlib import Path
import numpy as np

from polystore import DiskBackend, FileManager, MemoryBackend

registry = {
    "disk": DiskBackend(),
    "memory": MemoryBackend(),
}
files = FileManager(registry)

data = np.arange(6).reshape(2, 3)
files.save(data, Path("output.npy"), backend="disk")
loaded = files.load(Path("output.npy"), backend="disk")

For application startup where every discoverable context-free backend is wanted, use the package-owned lazy registry:

from polystore import FileManager, ensure_storage_registry, storage_registry

ensure_storage_registry()
files = FileManager(dict(storage_registry))

Backends requiring context-specific construction, including virtual workspaces and OMERO, must be instantiated by the application and added to its registry.

Nominal backend extension

Concrete backends inherit DataSource, DataSink, StorageBackend, or ReadOnlyBackend and declare their backend key. The authoritative class catalog is BackendBase.__registry__ (also exported as STORAGE_BACKENDS). Do not build a second backend class table.

Installation

python -m pip install polystore

The core install includes NumPy, ArrayBridge, metaclass-registry, file locking, image I/O, and Zarr/OME-Zarr support. Framework and streaming extras add their corresponding optional runtimes.

Full documentation: polystore.readthedocs.io