Quickstart

We’ll cover the following here:

Installation

1. Installation with fastembed

2. Installation without fastembed

Embedding, Inserting and Querying

1. add and query with fastembed

2. Qdrant without fastembed: Points, upsert and query

Installation

Installation with fastembed

Qdrant’s Python client ships with FastEmbed, an optional dependency for embedding text without handling

!pip install 'qdrant-client[fastembed]' --quiet

Initialize the Client

We support 3 production options:

1. Qdrant Cloud - Recommended for Getting Started

2. Qdrant Managed Deployment with your Cloud Provider - Recommendeded for Enteprises

3. Qdrant Self-Hosted with Docker - Recommended for those with specific requirements

In addition, Python client wraps numpy in a :memory: mode, which is useful for getting a feel of the client syntax.

from qdrant_client import QdrantClient

# client = QdrantClient(path="path/to/db")  # Persists changes to disk
# or
client = QdrantClient(":memory:")

Embedding, Inserting and Querying

We’ll use the FastEmbed library to embed text without handling the embedding model.

Embedding and Inserting

We wrap the Point creation and insertion into a single API: add for convenience.

# Prepare your documents, metadata, and IDs
docs = ["Qdrant has Langchain integrations", "Qdrant also has Llama Index integrations"]

client.add(
    collection_name="demo_collection",
    documents=docs,
)

['a3e23385a815464385a7589443f850db', 'd5bef7146f1541518cd767313f6569d5']

If you’d like to add points with control over the metadata, you can use the metadata parameter. Here is a quick example:

# Prepare your documents, metadata, and IDs
docs = ["Qdrant has Langchain integrations", "Qdrant also has Llama Index integrations"]
metadata = [
    {"source": "Langchain-docs"},
    {"source": "Linkedin-docs"},
]
ids = [42, 2]

# Use the new add method
client.add(
    collection_name="demo_collection",
    documents=docs,
    metadata=metadata,
    ids=ids
)

[42, 2]

Querying with text directly

At query time, we need to embed the incoming query and then search for the nearest neighbors. We can do this with the query API:

search_result = client.query(
    collection_name="demo_collection",
    query_text="This is a query document",
    limit=1
)
print(search_result)

[QueryResponse(id=42, embedding=None, metadata={'document': 'Qdrant has Langchain integrations', 'source': 'Langchain-docs'}, document='Qdrant has Langchain integrations', score=0.8276550115796268)]

Qdrant without fastembed

Collection

A collection is a set of points with the same dimensionality and a similarity metric (e.g. Dot, Cosine) defined on it. We can create a collection with the recreate_collection API:

from qdrant_client.http.models import Distance, VectorParams

client.recreate_collection(
    collection_name="test_collection",
    vectors_config=VectorParams(size=4, distance=Distance.DOT),
)

True

Notice that we didn’t have to explicitly create a collection with fastembed: If the collection exists already, we update or create a new one when using add.

Points

A point is a vector of floats with a unique identifier id. We can create a point with the Point API:

from qdrant_client.http.models import PointStruct

operation_info = client.upsert(
    collection_name="test_collection",
    wait=True,
    points=[
        PointStruct(id=1, vector=[0.05, 0.61, 0.76, 0.74], payload={"city": "Berlin"}),
        PointStruct(id=2, vector=[0.19, 0.81, 0.75, 0.11], payload={"city": "London"}),
        PointStruct(id=3, vector=[0.36, 0.55, 0.47, 0.94], payload={"city": "Moscow"}),
        PointStruct(id=4, vector=[0.18, 0.01, 0.85, 0.80], payload={"city": "New York"}),
        PointStruct(id=5, vector=[0.24, 0.18, 0.22, 0.44], payload={"city": "Beijing"}),
        PointStruct(id=6, vector=[0.35, 0.08, 0.11, 0.44], payload={"city": "Mumbai"}),
    ]
)
print(operation_info)

operation_id=0 status=<UpdateStatus.COMPLETED: 'completed'>

This creation of Points is also abstracted away with the add API.

Querying with Vector

search_result = client.search(
    collection_name="test_collection",
    query_vector=[0.18, 0.81, 0.75, 0.12],
    limit=1
)
print(search_result)

[ScoredPoint(id=2, version=0, score=1.2660000014305115, payload={'city': 'London'}, vector=None)]

Querying with a Filter and Vector

from qdrant_client.http.models import Filter, FieldCondition, MatchValue

search_result = client.search(
    collection_name="test_collection",
    query_vector=[0.2, 0.1, 0.9, 0.7],
    query_filter=Filter(
        must=[
            FieldCondition(
                key="city",
                match=MatchValue(value="London")
            )
        ]
    ),
    limit=1
)
print(search_result)

[ScoredPoint(id=2, version=0, score=0.8709999993443489, payload={'city': 'London'}, vector=None)]