agentic_vector library

Vector storage and similarity search for the agentic framework.

One VectorStore port, an exact in-process implementation, a Qdrant adapter, and the glue that turns an embedding model plus a store into something you hand text to.

import 'package:agentic_vector/agentic_vector.dart';

final index = EmbeddingIndex(
  model: OpenAiCompatibleEmbeddingModel.openAi(apiKey: key),
  store: InMemoryVectorStore(dimensions: 1536),
);

await index.addText(
  'Dart 3 added exhaustive pattern matching.',
  id: 'dart-3-patterns',
  metadata: {'topic': 'language'},
);

final hits = await index.query(
  'How do I match on a sealed type?',
  filter: MetadataFilter.equals('topic', 'language'),
  topK: 3,
);

Writing another backend is a class implementing VectorStore in your own package; nothing here needs to change. The MetadataFilter hierarchy is sealed on purpose, so the compiler tells you what your adapter has not translated yet.