Announcing Zvec v0.4.0

TL;DR: We are releasing Zvec v0.4.0, introducing the official Dart/Flutter SDK for Android and iOS. This release also expands the topK limit for large-scale recall; strengthens cross-platform stability on Windows and iOS.

You can find the complete release notes on GitHub.

Dart/Flutter SDK: Zvec Goes Mobile

Zvec v0.4.0 ships the first official Dart/Flutter SDK — bringing in-process, lightning-fast vector search to Android and iOS applications.

The package is published on pub.dev and can be added to any Flutter project with a single command:

flutter pub add zvec

The SDK is built on top of dart:ffi and covers synchronous vector operations. Prebuilt native libraries are automatically downloaded at build time, so end users require zero manual native compilation.

Platform support:

Quick start:

import 'package:zvec/zvec.dart';

Zvec.initialize();

final schema = CollectionSchema(name: 'demo', fields: [
  FieldSchema.vector('embedding', dimension: 128),
  FieldSchema.string('title'),
]);

final collection = Collection.createAndOpen('/path/to/db', schema);

// insert, query, fetch ...

collection.close();
Zvec.shutdown();

Enlarged topK Limit

The hard upper bound on topK results has been relaxed. You can now retrieve significantly larger result sets in a single query, enabling use cases such as large candidate recall followed by downstream re-ranking.

import zvec

result = collection.query(
  vectors=zvec.VectorQuery(
    field_name="dense_embedding",
    vector=[0.1] * 768,
  ),
  topk=70000, # Retrieve a massive batch of results
  include_vector=False,
)

print(result)

Improvements & Fixes

• SQ8 Quantizer Recall: Fixed a recall degradation caused by missing int8 rounding in SQ8 quantizer metadata computation.
• Windows Compatibility: Fixed drive-root path handling in MakePath and added GetLastErrorString() for more accurate OS-level error reporting.
• Collection Path: Relaxed overly strict collection storage path validation, enabling more flexible deployment configurations.

Roadmap

For a glimpse into our future plans — including storage scalability, expanded algorithm support, and more language SDKs — please visit our official roadmap.