zvec_flutter

Flutter Android FFI wrapper around the bundled zvec-android-arm64 build.

This package follows the same approach used in the local edgemind app: a small C-compatible wrapper over the zvec C++ API, then a Dart dart:ffi surface for Flutter.

Scope

• Android only
• arm64-v8a only
• Default schema: id, vector, content, metadata, hash
• Default vector index: HNSW with inner product

Python-Like Dart API

final schema = CollectionSchema(
  name: 'example',
  vectors: const VectorSchema(
    'embedding',
    DataType.VECTOR_FP32,
    4,
    indexParam: HnswIndexParam(metricType: MetricType.IP),
  ),
);

final collection = create_and_open(
  path: '/data/user/0/your.app/files/zvec_demo',
  schema: schema,
);

collection.insert([
  Doc(
    id: 'doc_1',
    vectors: {
      'embedding': [0.1, 0.2, 0.3, 0.4],
    },
    fields: {
      'content': 'Flutter talks to zvec through FFI.',
      'hash': 'doc_1',
    },
  ),
]);

final results = collection.query(
  const VectorQuery('embedding', vector: [0.4, 0.3, 0.3, 0.1]),
  topk: 10,
);

Notes

• The repository tracks the final prebuilt arm64-v8a JNI bridge libzvec_flutter_android.so only.
• The Android build copies libc++_shared.so from Android NDK 28.2.13676358 at build time.
• Maintainers can rebuild the native bridge from source in this repository by passing -PzvecFlutterBuildFromSource=true to the Android build.
• The example app in example/ shows end-to-end usage with local storage and search.
• The high-level Dart API follows the upstream Python naming and concepts where practical, but the current Android wrapper supports one primary dense vector field and stores extra scalar fields in JSON metadata.
• If you want to regenerate low-level bindings later, point ffigen at src/zvec_c.h.