Installation

Version scope: Python 0.0.12 / JavaScript 0.5.4. Entity keys, progressive geometry, structured ragged/CSR and additional index capabilities are included in these releases. See feature examples and limits for the supported boundaries; older packages may not expose these methods.

DreamDB ships two SDKs, both compiled from the same Rust core, so they agree byte for byte. Python is the ingestion and training surface. JavaScript runs in Node and the browser, reads everywhere, and writes where you give it a backend that can PUT.

	Python	JavaScript
Package	`dreamdb`	`@dreamlake/dreamdb`
Runs on	Python 3.8+	Node 20+ and modern browsers
Reads	Yes	Yes
Writes	Yes	Yes, from 0.4.0
Creates datasets	Yes	Node only, and not for IVF or IMI index families
Trains indexes	Yes	No — train in Python or the CLI, attach as a layer

If you are only searching an existing dataset, the JavaScript SDK alone is enough. If you are producing data at any scale, or training an index, you want Python.

Python

bash

pip install dreamdb==0.0.12 numpy

numpy is optional but every example uses it. With uv:

bash

uv add dreamdb==0.0.12 numpy

Check it:

bash

python -c "from importlib.metadata import version; print(version('dreamdb'))"
# 0.0.12

The wheel bundles a compiled Rust extension. PyPI publishes CPython 3.8 ABI-stable wheels for macOS on Apple silicon and Linux on x86_64 and ARM (manylinux2014). There is no source distribution or prebuilt Windows/Intel Mac wheel in 0.0.12, so those platforms need a source checkout, a Rust toolchain, and maturin; pip install dreamdb alone cannot build a source archive that PyPI does not publish.

JavaScript

bash

npm install @dreamlake/dreamdb@0.5.4

Check it:

bash

node -e "import('@dreamlake/dreamdb').then(m => console.log(typeof m.Space))"
# function

The package is a WebAssembly build of the same Rust core the Python SDK uses. Shared implementation is not an independent interoperability proof, and target capabilities differ. It ships three entry points and picks one for you:

Import	Resolves to	Notes
`@dreamlake/dreamdb` in a bundler	bundler build	Vite needs `vite-plugin-wasm`
`@dreamlake/dreamdb` in Node	Node build	ESM and CJS both work. The only build with `Authoring`
`@dreamlake/dreamdb/web`	web build	No bundler plugin. Call `await ready()` before anything else

html

<script type="module">
  import ready, { Space } from "https://esm.sh/@dreamlake/dreamdb/web";
  await ready();
</script>

The browser build leaves out dataset creation and index building. Reading, querying, appending and committing are present; Authoring throws in the browser with an explanation. Artifact size depends on the version and build toolchain; historical bundle sizes are not a current download-size promise.

Collector upgrade: 0.0.12 repairs GC retention for reference-mode vector data. Upgrade or fence all collectors before using GC on those roots, including retained history. A reader-only upgrade does not protect against an old collector. See upgrade details before scheduling destructive maintenance.

The pure-TypeScript package that used to carry this name (0.1.x) is retired. Its writes produced objects no other implementation could read, and its own tests passed because they were its reader reading its writer. 0.2.x and 0.3.x were read-only while the Rust core was ported; 0.4.0 restores writing.

Installation

Python

JavaScript

What to read next