Create Collection

Learn how to create collections in Qdrant with various vector configurations including dense vectors, sparse vectors, and named vectors.

A collection is a named set of points (vectors with a payload) that you can search. Creating a collection is the first step before inserting data into Qdrant.

API Endpoint

PUT /collections/{collection_name}

Create Collection with Dense Vectors

1. Define collection parameters

   Create a collection with a single dense vector configuration. This is the most common use case.

2. Send the request

   Use the REST API or Python client to create the collection.

REST API

curl -X PUT http://localhost:6333/collections/my_collection \
  -H 'Content-Type: application/json' \
  -d '{
    "vectors": {
      "size": 384,
      "distance": "Cosine"
    }
  }'

Vector Configuration Parameters

vectors.size · integer · required

Size of the vector (number of dimensions).

vectors.distance · string · required

Distance metric for vector comparison. Available options:

• Cosine - Cosine similarity
• Euclid - Euclidean distance
• Dot - Dot product
• Manhattan - Manhattan distance

vectors.hnsw_config · object

HNSW index configuration for fast approximate nearest neighbor search.

vectors.quantization_config · object

Quantization configuration to reduce memory usage.

vectors.on_disk · boolean · default: "false"

Store vectors on disk instead of RAM to save memory.

Create Collection with Named Vectors

Use named vectors when you need multiple vector representations per point (e.g., text and image embeddings).

REST API

curl -X PUT http://localhost:6333/collections/multi_vector_collection \
  -H 'Content-Type: application/json' \
  -d '{
    "vectors": {
      "text": {
        "size": 768,
        "distance": "Cosine"
      },
      "image": {
        "size": 512,
        "distance": "Euclid"
      }
    }
  }'

Create Collection with Sparse Vectors

Sparse vectors are useful for keyword-based search and can be combined with dense vectors for hybrid search.

REST API

curl -X PUT http://localhost:6333/collections/sparse_collection \
  -H 'Content-Type: application/json' \
  -d '{
    "vectors": {
      "dense": {
        "size": 384,
        "distance": "Cosine"
      }
    },
    "sparse_vectors": {
      "text": {}
    }
  }'

Additional Configuration Options

Optimizers Config

Control how Qdrant optimizes storage and indexing.

REST API

curl -X PUT http://localhost:6333/collections/optimized_collection \
  -H 'Content-Type: application/json' \
  -d '{
    "vectors": {
      "size": 384,
      "distance": "Cosine"
    },
    "optimizers_config": {
      "indexing_threshold": 20000,
      "memmap_threshold": 50000
    }
  }'

Replication Factor

For distributed deployments, specify how many replicas to create.

REST API

curl -X PUT http://localhost:6333/collections/replicated_collection \
  -H 'Content-Type: application/json' \
  -d '{
    "vectors": {
      "size": 384,
      "distance": "Cosine"
    },
    "replication_factor": 2,
    "write_consistency_factor": 1
  }'

Replication factor must be less than or equal to the number of nodes in your cluster.

Response Format

Successful collection creation returns:

result · boolean

Indicates whether the operation was successful.

status · string

Operation status, typically “ok” on success.

time · number

Time taken to execute the operation in seconds.

Response Example

{
  "result": true,
  "status": "ok",
  "time": 0.031095
}

Query Parameters

timeout · integer

Wait timeout in seconds for the operation to complete. If timeout is reached, the service will return an error.

Collection names must be unique. Creating a collection with an existing name will return an error.

Check if Collection Exists

Before creating a collection, you can check if it already exists:

GET /collections/{collection_name}/exists

REST API

curl http://localhost:6333/collections/my_collection/exists