Pinecone Index Management

An index defines the vector storage configuration, including dimension, metric and deployment specification.

#Serverless and legacy pods

The source contrasts usage-oriented serverless resources with reserved pod-based capacity and records restrictions for newer accounts after August 2025. Confirm current account eligibility rather than treating a legacy tutorial as a product catalog.

#Core configuration

Dimension must match the embedding output; it is not a single universal number for all OpenAI or other embedding models. Common metrics include cosine, Euclidean and dot product. Choose cloud/region deliberately.

For an integrated-embedding index, also check the supported embedding model and field mapping. That determines whether the ingestion path supplies vectors or source text.

#Lifecycle examples

pythonCopy
print(pc.list_indexes())
description = pc.describe_index("my-index")
print(description.status["ready"])
pc.configure_index("my-index", deletion_protection="enabled")
# Destructive; review the target and backup requirements first:
# pc.delete_index("my-index")

pythonCopy
index = pc.Index("my-index")
stats = index.describe_index_stats()
print(stats.total_vector_count)

These fragments assume an authenticated client and an existing index.

#Operational design

Place application and index in compatible nearby regions where appropriate. Keep metadata focused on source IDs, URLs and useful filters rather than unnecessary large payloads. Plan the supported metadata-filter schema and retention policy before the corpus becomes difficult to migrate.

#Next step

• Namespaces