a minimal & modular Zarr implementation in TypeScript

• Zero dependencies (optionally scijs/ndarray)
• Runs natively in Node, Browsers, and Deno (ESM)
• Supports v2 or v3 protocols, C & F-order arrays, diverse data-types, and ZEP2 Sharding
• Allows flexible storage backends and compression codecs
• Provides rich, in-editor type information via template literal types

zarrita supports a variety of environments, including the browser, Node.js, and Deno.

If you're developing an application with Node.js, you can install zarrita with yarn, npm, or pnpm:

npm install zarrita@next

Read the documentation to learn more about other environments.

import * as zarr from "zarrita";

const store = new zarr.FetchStore("http://localhost:8080/data.zarr");
const arr = await zarr.open(store, { kind: "array" }); // zarr.Array<DataType, FetchStore>

// read chunk
const chunk = await arr.getChunk([0, 0]);

// Option 1: Builtin getter, no dependencies
const full = await zarr.get(arr); // { data: Int32Array, shape: number[], stride: number[] }

// Option 2: scijs/ndarray getter, includes `ndarray` and `ndarray-ops` dependencies
import { get } from "@zarrita/ndarray";
const full = await get(arr); // ndarray.Ndarray<Int32Array>

// read region
const region = await get(arr, [null, zarr.slice(6)]);

Read the documentation to learn more.

zarrita's API is almost entirely tree-shakeable, meaning developers are able to pick and choose the features necessary for an application. This design choice differs from existing implemenations of Zarr in JavaScript, and allows zarrita to be both minimal and more feature-complete if necessary.

classDiagram
    indexing --|> core : uses
    ndarray --|> indexing : uses
    ndarray --|> core : uses
    core --|> storage : uses

    class indexing {
        - get(arr: zarr.Array, selection)
        - set(arr: zarr.Array, selection, view)
        - slice and index multiple chunks
        - returns strided arrays
    }

    class ndarray {
        - get(arr: zarr.Array, selection)
        - set(arr: zarr.Array,  selection, view)
        - slice and index multiple chunks
        - returns scijs/ndarray objects
    }

    class core {
        - open(store: Readable)
        - create(store: Writable)
        - zarr.Array and zarr.Group
        - access and decode individual chunks
    }

    class storage {
        - Readable
        - Writable
        - Map()
        - FetchStore()
        - FileSystemStore()
        - ReferenceStore()
        - ZipStore()
    }

This library uses the pnpm package manager.

pnpm install
pnpm build
pnpm test

The tests are run with Vitest, which is a Node.js test runner. To try our a development version of zarrita in the browser, run pnpm build and start a web-server in the root of the repository:

python3 -m http.server .
# navigate to localhost:8000/demo.html

You can edit the contents of demo.html and refresh the page.