npm / @pinecone-database/pinecone
@pinecone-database/pinecone
What changed in @pinecone-database/pinecone 8 to 9, from its official release notes, upgrade guides and advisories. Sources checked 2026-10-18.
19 breaking1 security4 deprecations15 new APIs and options3 behavior and defaults1 tooling3 fixes and other
See which of these apply to your code, ranked, in your agent:
npx @unstale/cli init@pinecone-database/pinecone 938 changes
Breaking changes 19
Use BackupListPagination instead of the removed BackupPaginationResponse type.
createdAt now decodes to an optional Date; use toISOString instead of string methods.
Backup dimension and metric fields move into the optional schema.
Document update and delete now return response objects instead of void, which may require updating explicitly typed wrappers.
Top-level configuration embed is rejected; update semantic fields through a schema.fields patch or create an integrated-embedding index with createForModel.
Pod creation, legacy metadata-indexing schemas, and sourceCollection/sourceBackupId creation options are rejected; use the backup restoration resource method for restoration.
Do not mix legacy dimension, metric, vectorType, or spec fields with native schema or deployment fields in one creation request.
recordCount and sizeBytes are now optional decimal strings instead of optional numbers; convert explicitly for arithmetic and handle missing values.
Node.js 22 or later is required to use v9.
next is required when a pagination object exists; omit an empty pagination envelope or provide a real cursor.
Use this resource method to restore an index from a backup.
Preview document methods are removed; migrate document operations to the index namespace documents accessor and stable request and response types.
Preview exports are removed; migrate preview request and response types to their stable equivalents.
Use ReadCapacityDedicated instead of ReadCapacityDedicatedSpec.
Response declarations now require nodeType, scaling, and manual replicas and shards; capacity status counts can be null.
Use ReadCapacityOnDemand instead of ReadCapacityOnDemandSpec.
Requests reject resource identifiers whose dot segments would resolve to a different URL path.
Timestamps and several response fields can now be null or absent, so check values before accessing them.
SearchMatchTerms now requires both strategy and terms; omit optional matchTerms when no filter is intended.
Security 1
Request-header logging redacts Api-Key and Authorization across header casing and supported header container formats.
Deprecations 4
Flat control methods remain available as deprecated delegates; use pc.indexes.configure for new code.
Use pc.backups.create instead of the deprecated flat control method.
Use pc.backups.createIndex instead of the deprecated flat control method.
Use pc.backups.listByIndex or pc.backups.list instead of the deprecated flat control method.
New APIs and options 7
Assistant uploads accept Uint8Array, Blob, and async iterables of byte or string chunks, including Node Buffer and Readable inputs; iterable uploads require Node.js and are sent once.
The main client now provides document upsert, search, fetch, update, delete, and list operations through the index namespace documents accessor.
HTTP 412 errors now have a dedicated error class.
HTTP 402 errors now have a dedicated error class.
Legacy properties derive values from schema and deployment, are non-enumerable, and can be undefined or throw when values are unavailable or ambiguous.
Create and manage daily, weekly, or monthly backup schedules with retention periods and inspect their run history.
Define searchable fields when creating an index for schema-based document operations.
Behavior and default changes 3
Client additionalHeaders now propagate to index and assistant clients, including upload and streaming calls; target-specific exact-key matches take precedence.
Client construction now uses globalThis.fetch, and importing the package no longer eagerly loads Node built-ins.
An exact X-Pinecone-Api-Version header key now overrides the default API version.
Tooling 1
Published declarations support TypeScript 5.2 through 7.x, including CommonJS and ESM consumer configurations.
Fixes and other changes 3
Chat streams now preserve multibyte UTF-8 characters split across byte chunks.
Operation Content-Type remains protected regardless of header casing.
Nested API error envelopes now expose the server's message correctly.
@pinecone-database/pinecone 88 changes
New APIs and options 8
Adds a top-level client for managing organizations, projects, API keys, service accounts, role bindings, invites, and users through the Pinecone Admin API.
Uses service-account OAuth2 instead of a project
Api-Key; fetches and caches the bearer token on the first admin request, with no proactive refresh.Adds an
initAdmin()helper tonpm run replfor spinning up anAdminClient.Can be set in the environment to provide the service-account client ID used by
AdminClient.Can be set in the environment to provide the service-account client secret used by
AdminClient.Adds an n-gram tokenizer option for full-text search string fields, enabling substring matching; it cannot be combined with
stemmingorstopWords.Adds an alpha preview method to list document IDs for a namespace, with prefix filtering, a result limit, and cursor-based pagination.
Adds an alpha preview method for partial updates to existing documents by
_id, including setting values or removing fields; updates to nonexistent documents are accepted as a no-op.