@galdor/memory-qdrant

A Qdrant-backed long-term memory store for galdor, implementing the core memory.Store interface over the Qdrant HTTP REST API. It drops in behind a Retriever like the bundled InMemoryStore, but persists vectors in a Qdrant collection and runs the nearest-neighbour search server-side. This backend is vector-only (retrieve requires query.embedding).

Usage

import { openQdrant } from "@galdor/memory-qdrant";

const store = await openQdrant({
  url: "http://localhost:6333",   // Qdrant HTTP API
  collection: "galdor_chunks",    // optional; created if missing
  dim: 1024,                      // embedding dimensionality
  apiKey: process.env.QDRANT_API_KEY, // optional (Qdrant Cloud / auth)
});

await store.add([{ id: "c1", documentId: "d1", index: 0, text: "…", embedding: vec, metadata: { lang: "es" } }]);
const hits = await store.retrieve({ embedding: queryVec, k: 5, filter: { lang: "es" } });
await store.delete("d1");
await store.close();

Behavior

• open creates the collection (Cosine distance, the given dim) if missing.
• add upserts by chunk id (a deterministic UUID-shaped point id), storing documentId/index/text + metadata in the payload. Rejects empty ids, dimension mismatches, and reserved __-prefixed metadata keys.
• retrieve returns top-K by Cosine score (higher = better; negatives dropped); filter maps to a Qdrant must exact-match clause.
• delete removes every point whose __document_id matches.