Pinecone allows you to partition the records in an index into namespaces. Queries and other operations are then limited to one namespace, so different requests can search different subsets of your index.

For example, you might want to define a namespace for indexing articles by content, and another for indexing articles by title. For a complete example, see our Namespaces example notebook.

Every index is made up of one or more namespaces. Every record exists in exactly one namespace.

Namespaces are uniquely identified by a namespace name, which almost all operations accept as a parameter to limit their work to the specified namespace. When you don’t specify a namespace name for an operation, Pinecone uses the default namespace name of "" (the empty string).

Creating a namespace

Namespaces are created automatically the first time they are used to upsert records. If the namespace doesn’t exist, it is created implicitly.

The example below will create a "my-first-namespace" namespace if it doesn’t already exist:

# Upsert records while creating a new namespace
index.upsert(vectors=[('id-1', [0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1])],

Then you can submit queries and other operations specifying that namespace as a parameter. For example, to query the records in namespace "my-first-namespace" use code like the following:


index.query(vector=[0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1], top_k=1, namespace='my-first-namespace')

const ns1 = index.namespace('my-first-namespace'); await ns1.query({ topK: 1, vector: [0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1], });

INDEX_NAME='index-name' \ PROJECT='project-id' \ ENVIRONMENT='your-environment' \ API_KEY='your-api-key' \ curl -X POST "https://$INDEX_NAME-$PROJECT.svc.$" \ -H "Api-Key: $API_KEY" \ -H 'Content-Type: application/json' \ -d '{ "topK": 1, "vector": [0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1], "namespace": "my-first-namespace" }'

Creating more than one namespace

You can create more than one namespace. For example, insert data into separate namespaces:

import numpy as np

# Create three sets of 8-dimensional vectors
vectors_a = np.random.rand(15, 8).tolist()
vectors_b = np.random.rand(20, 8).tolist()
vectors_c = np.random.rand(30, 8).tolist()

# Create ids
ids_a = map(str, np.arange(15).tolist())
ids_b = map(str, np.arange(20).tolist())
ids_c = map(str, np.arange(30).tolist())

# Insert into separate namespaces

# If no namespaces are specified, the index uses the default namespace ''

# At this point, index.describe_index_stats() returns:
# {'dimension': 8,
#  'namespaces': {'': {'vector_count': 30},
#                 'namespace_a': {'vector_count': 15},
#                 'namespace_b': {'vector_count': 20}}}

Operations across all namespaces

All vector operations apply to a single namespace, with one exception:

The DescribeIndexStatistics operation returns per-namespace statistics about the contents of all namespaces in an index. More details