What this operation supports in Pinecone
Inserts a new vector or overwrites an existing one that shares the same ID, including its embedding values and optional metadata.
Implementation details
Because this is an upsert, sending a vector with an ID that already exists replaces its previous embedding and metadata entirely rather than merging them.
Operational considerations
Keeping embedding dimensions consistent with the index's configured dimension is required -- a mismatched vector size causes the upsert to fail.
Use cases
Pinecone Upsert enables data freshness in automated pipelines. This operation is essential when embeddings are regenerated or corrected and must be reflected immediately in downstream workflows.