Qdrant

Qdrant is a vector database for similarity search. memio's Qdrant adapter implements the DocumentStore protocol using qdrant-client's built-in fastembed integration for automatic embedding generation.

Installation

pip install memio[qdrant]

This installs qdrant-client[fastembed], which includes the fastembed library for local embedding generation.

Setup

Qdrant supports multiple deployment modes. You create a client and pass it to the adapter:

In-memory (no persistence, great for testing)

from qdrant_client import AsyncQdrantClient
from memio import Memio
from memio.providers.qdrant import QdrantDocumentAdapter

client = AsyncQdrantClient(":memory:")
memio_client = Memio(
    documents=QdrantDocumentAdapter(client=client, collection_name="my-docs"),
)

On-disk (persistent, no server needed)

client = AsyncQdrantClient(path="./qdrant_data")

Qdrant Cloud

client = AsyncQdrantClient(
    url="https://your-cluster.cloud.qdrant.io:6333",
    api_key="your-api-key",
)

Self-hosted server

client = AsyncQdrantClient(host="localhost", port=6333)

Supported stores

Store	Adapter
DocumentStore	QdrantDocumentAdapter

Usage

import asyncio
from qdrant_client import AsyncQdrantClient
from memio import Memio
from memio.providers.qdrant import QdrantDocumentAdapter

async def main():
    client = AsyncQdrantClient(":memory:")
    memio_client = Memio(
        documents=QdrantDocumentAdapter(client=client, collection_name="docs"),
    )

    # Add a document
    doc = await memio_client.documents.add(
        content="Qdrant is a vector database for similarity search",
        metadata={"source": "docs"},
    )

    # Search
    results = await memio_client.documents.search(query="vector database", limit=5)
    for r in results:
        print(f"{r.content} (score: {r.score})")

    # Search with metadata filter
    results = await memio_client.documents.search(
        query="vector", filters={"source": "docs"},
    )

    # Get all documents
    all_docs = await memio_client.documents.get_all()

    # Get by ID
    retrieved = await memio_client.documents.get(doc_id=doc.id)

    # Update
    updated = await memio_client.documents.update(
        doc_id=doc.id, content="Updated content",
    )

    # Delete
    await memio_client.documents.delete(doc_id=doc.id)

    # Delete all
    await memio_client.documents.delete_all()

    await client.close()

asyncio.run(main())

Embedding model

By default, Qdrant uses the BAAI/bge-small-en embedding model (384 dimensions) via fastembed. The model is downloaded automatically on first use.

To use a different model:

client = AsyncQdrantClient(":memory:")
client.set_model("sentence-transformers/all-MiniLM-L6-v2")

Known quirks

• Auto-collection creation: Collections are created automatically on first add() call. No manual setup needed.
• UUID IDs: When no doc_id is provided, a UUID is generated automatically.
• Scores: Search scores are returned directly from Qdrant's similarity scoring (higher = more similar).
• Metadata filtering: The filters parameter in search() and get_all() supports exact-match filtering on payload fields.