qdrant_edge_flutter library

On-device Qdrant vector search for Flutter, powered by the qdrant-edge Rust engine over dart:ffi. Everything runs locally — no server, no network.

QdrantEdge is the client/factory: it creates Shards (the vector store) and the on-device embedders (Bm25 sparse, Dense semantic). Complex arguments are plain Dart Map/List that are passed to the engine as JSON, mirroring the Qdrant request model.

final client = QdrantEdge();
final shard = client.createShard('${dir.path}/notes', {
  'vectors': {'dense': {'size': 384, 'distance': 'Cosine'}},
});
shard.upsert([
  {'id': 1, 'vector': {'dense': embedding}, 'payload': {'title': 'fox'}},
]);
final hits = shard.search({'vector': query, 'using': 'dense', 'limit': 5});
shard.close();

For the simple "add text / search text" workflow, see TextIndex via QdrantEdge.openTextIndex.

Classes

Bm25

On-device BM25 sparse embedder. Reusable across texts and shards.

Dense

On-device dense (semantic) embedder. Reusable across texts and shards.

QdrantEdge

The client / factory. Cheap to construct; the native library loads once.

Shard

An open vector shard — the on-device store + index. Operations are synchronous; for large batches run them in an isolate. Unusable after close.

TextIndex

Convenience wrapper for text-first use: it owns a Shard plus a Bm25 embedder (and an optional Dense one) and embeds text for you. Lexical-only by default; pass modelDir to QdrantEdge.openTextIndex for hybrid.

Exceptions / Errors

QdrantEdgeException

Thrown when a native call fails. The message comes from Rust.