Qdrant Python Client

1.17.0 · active · verified Sat Feb 28

Official Python client for the Qdrant vector search engine. Supports both remote server (gRPC or REST) and local in-memory/on-disk mode without a running server. All data operations use query_points() as the unified interface as of 1.10+. Prior search/recommend/discover methods removed in 1.14.0. Bundles FastEmbed for optional local embedding generation. Async support via AsyncQdrantClient.

Warnings

breaking client.search(), client.recommend(), client.discover(), client.search_batch(), client.recommend_batch(), client.discovery_batch() all removed in 1.14.0. Massive amount of tutorials, LangChain/LlamaIndex integrations, and LLM-generated code uses client.search(). Raises AttributeError on 1.14+.
Fix: Replace all with client.query_points(). For batch: client.query_batch_points(). Migration guide: qdrant.tech/documentation/concepts/search/
breaking client.upload_records() removed in 1.14.0. Raises AttributeError.
Fix: Replace with client.upload_points() using PointStruct(id=..., vector=..., payload=...) objects.
breaking init_from parameter in create_collection() removed in 1.14.0.
Fix: Use client.update_collection_aliases() or migrate data manually if you were using init_from for collection cloning.
breaking Pydantic <2.2.1 is not supported. Supported versions are v1.10.x and >=2.2.1. Intermediate Pydantic 2.0–2.2.0 raises validation errors.
Fix: Pin pydantic to either >=1.10,<2.0 or >=2.2.1. Avoid 2.0.0–2.2.0.
gotcha Local in-memory mode (QdrantClient(':memory:')) does not persist. Data is lost when the process exits. QdrantClient(path='...') persists to disk but does not support concurrent multi-process access.
Fix: For production or multi-process access, run a Qdrant server (Docker or binary) and connect via QdrantClient(url='http://localhost:6333').
gotcha collection.query_points() offset parameter is no longer propagated into prefetches for nested queries. Pre-1.13 behavior silently passed offset into sub-queries; current behavior does not.
Fix: Apply offset only at the top-level query_points() call, not in nested prefetch structures.
gotcha AsyncQdrantClient raw gRPC methods are not accessible from the sync QdrantClient as of 1.11. Mixing sync client with raw async gRPC calls raises AttributeError.
Fix: Use AsyncQdrantClient for async gRPC operations. Do not attempt to access .grpc_* methods from the synchronous QdrantClient.

Install

pip install qdrant-client Base (REST + gRPC + local mode)
pip install 'qdrant-client[fastembed]' With FastEmbed for local embeddings

Imports

query_points (unified search API)
```
from qdrant_client import QdrantClient
from qdrant_client.models import Distance, VectorParams, PointStruct, QueryRequest
client = QdrantClient(':memory:')
results = client.query_points(collection_name='my_col', query=[0.1, 0.2], limit=5)
```
client.search(), client.recommend(), client.discover(), client.search_batch(), client.recommend_batch(), client.discovery_batch() all removed in 1.14.0. Raises AttributeError. Replace all with client.query_points() — the unified interface introduced in 1.10.
upload_points (replaces upload_records)
```
client.upload_points(collection_name='my_col', points=[PointStruct(id=1, vector=[0.1, 0.2], payload={'text': 'hello'})])
```
upload_records() removed in 1.14.0. Replace with upload_points() using PointStruct objects.

Quickstart

query_points() is the unified search interface in 1.10+. upload_points() replaces the removed upload_records().

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

# In-memory (no server needed)
client = QdrantClient(':memory:')

# Persistent local mode
# client = QdrantClient(path='/path/to/db')

# Remote server
# client = QdrantClient(url='http://localhost:6333')

client.create_collection(
    collection_name='my_docs',
    vectors_config=VectorParams(size=4, distance=Distance.COSINE),
)

client.upload_points(
    collection_name='my_docs',
    points=[
        PointStruct(id=1, vector=[0.1, 0.2, 0.3, 0.4], payload={'text': 'hello'}),
        PointStruct(id=2, vector=[0.5, 0.6, 0.7, 0.8], payload={'text': 'world'}),
    ],
)

results = client.query_points(
    collection_name='my_docs',
    query=[0.1, 0.2, 0.3, 0.4],
    limit=2,
)
print(results.points)

view raw JSON →