// the take
The official Node/TypeScript client for Pinecone's hosted vector database, covering vector upsert/query, the newer document-oriented full-text + semantic search API, index/backup management, and the separate Assistant (RAG-as-a-service) API. It's for TypeScript developers already using or evaluating Pinecone as their managed vector store, not for anyone wanting a local or open-source vector DB.
Strong typing throughout, including generic metadata types for vector/document payloads, which catches a real class of bugs in filter and upsert code. The integrated embedding feature (declare `embed: { model }` on a field and Pinecone handles embedding on upsert and query) removes an entire category of client-side plumbing. CI is genuinely broad — it compiles a consumer project against TypeScript 5.2 through 7.x in both CommonJS and ESM, and runs against Node 22.x and 24.x, so declaration-file breakage gets caught before release. Error handling has dedicated types per failure mode (config, request, HTTP, batch) rather than one generic exception.
It's a thin client over a proprietary, paid backend — there's no self-hosted Pinecone, so adopting this SDK means adopting the vendor and its pricing, not just a library. The presence of four migration guides (v1, v2, v8, v9) points to a history of breaking API changes; the legacy pod-index accessors, `createForModel`, and the old Records API are all still carried alongside the new document/full-text API, which is real ongoing complexity for anyone reading the source rather than just the docs. Node >=22 as a hard minimum will knock out projects still on Node 20 LTS. The SDK explicitly can't be used safely from a browser (API key exposure), so anything client-side needs a backend proxy layer you have to build yourself — that's expected for vector DB clients but worth flagging as an operational requirement, not a given.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →