Qdrant

What Is Qdrant?

Qdrant is a vector database built around similarity search. It stores vectors together with payload data and lets applications retrieve the most relevant points using vector similarity and filters.

Qdrant vs FAISS

FAISS and Qdrant can both help you perform vector similarity search, but they solve different layers of the problem.

Core Qdrant Concepts

How a Qdrant Collection Looks

Keeping metadata with the vector is important because retrieval usually needs more than a numerical score. Your application may need the document title, source URL, tenant, permissions, or other fields before showing or generating an answer.

Python: Connect to Qdrant

For local development, Qdrant can run as a service and the Python client can connect to it. The exact URL depends on how you deploy Qdrant.

from qdrant_client import QdrantClient

client = QdrantClient(
    url="http://localhost:6333"
)

print(client.get_collections())

Create a Collection

A collection defines how Qdrant stores and searches vectors. One important setting is the vector size, which must match the output dimension of your embedding model.

from qdrant_client import QdrantClient
from qdrant_client.models import Distance, VectorParams

client = QdrantClient(url="http://localhost:6333")

client.create_collection(
    collection_name="documents",
    vectors_config=VectorParams(
        size=384,
        distance=Distance.COSINE,
    ),
)

print("Collection created")

Insert Vectors with Payloads

Qdrant points can contain an ID, vector, and payload. Payloads let retrieval keep useful structured information close to the embedding.

from qdrant_client.models import PointStruct

points = [
    PointStruct(
        id=1,
        vector=[0.10, 0.20, 0.30, 0.40],
        payload={
            "title": "Password Reset",
            "category": "support",
            "language": "en",
        },
    ),
    PointStruct(
        id=2,
        vector=[0.80, 0.10, 0.05, 0.20],
        payload={
            "title": "Shipping Policy",
            "category": "commerce",
            "language": "en",
        },
    ),
]

client.upsert(
    collection_name="documents",
    points=points,
)

print("Points inserted")

Search for Similar Vectors

A query vector can be compared against vectors in a collection. Qdrant returns the most relevant points according to the collection's configured distance metric.

query_vector = [0.12, 0.18, 0.28, 0.42]

results = client.query_points(
    collection_name="documents",
    query=query_vector,
    limit=2,
).points

for result in results:
    print(result.id, result.score, result.payload)

Metadata Filtering

Semantic similarity alone is often not enough. A search may need to find content that is both semantically relevant and allowed by an application rule.

from qdrant_client.models import FieldCondition, Filter, MatchValue

query_vector = [0.12, 0.18, 0.28, 0.42]

results = client.query_points(
    collection_name="documents",
    query=query_vector,
    query_filter=Filter(
        must=[
            FieldCondition(
                key="category",
                match=MatchValue(value="support"),
            )
        ]
    ),
    limit=3,
).points

for result in results:
    print(result.payload["title"], result.score)

Why Payloads Matter in RAG

For example, a multi-tenant support assistant may need to retrieve semantically relevant chunks while ensuring that the chunks belong to the correct customer. Payload filters can participate in that retrieval design.

Qdrant in a Semantic Search Pipeline

The embedding model creates the vector representation. Qdrant handles storage and retrieval. The application decides what to do with the retrieved records.

Qdrant Local vs Managed Deployment

Qdrant vs a Traditional SQL Database

These systems are not necessarily competitors. A production application may use both: SQL for transactional business data and Qdrant for semantic retrieval.

Common Mistakes

When Should You Use Qdrant?