Pinecone

Pinecone

This connector calls the Pinecone API (version 2026-07) with the user's API key: indexes and their statistics, semantic and keyword search, vector queries, fetching and listing records, namespaces, Pinecone Inference (embedding and reranking) and Pinecone Assistant chat. With the user's confirmation it can add text records and change metadata.

Setup (about 2 minutes)

1. Open the Pinecone console (https://app.pinecone.io), pick the project, open API keys and click + API key.
2. Choose the permissions. For read-only use pick ProjectViewer (or ControlPlaneViewer plus DataPlaneViewer); search, embedding, reranking and assistant chat all work with it. Add write access (ProjectEditor or DataPlaneEditor) only if the AI may write records. Starter and Builder plans only offer All.
3. Copy the key (Pinecone shows it once, it usually starts with pcsk_) into PINECONE_API_KEY. It is sent as the Api-Key header with X-Pinecone-Api-Version: 2026-07. One key covers one project.

Index hosts: every index has its own data host, e.g. my-index-abc1234.svc.aped-4627-b74a.pinecone.io. Call pinecone_list_indexes or pinecone_describe_index first and pass their index_host value (the host without .pinecone.io) to every data tool. The connector only ever sends the key to *.pinecone.io.

Which tool

• pinecone_list_indexes, pinecone_describe_index: names, index_host, schema (dimension, metric, embedding model) and status. pinecone_describe_index_stats and pinecone_list_namespaces give record counts.
• Text search depends on the index type, which the schema shows:
  • integrated index (a semantic_text field, created with a hosted model): pinecone_search_records with plain text;
  • document index (typed string fields): pinecone_search_documents with text (keyword) or embed (semantic) clauses;
  • index of the user's own vectors (_values field): pinecone_query_vectors by record id, or with a vector from pinecone_embed using the same model as the index.
• pinecone_fetch_vectors and pinecone_list_vector_ids read records by id; vector values are never passed on.
• pinecone_list_models, pinecone_embed, pinecone_rerank: Pinecone Inference.
• pinecone_list_assistants, pinecone_chat_assistant: questions to a Pinecone Assistant over its uploaded files, with citations.

Writes: pinecone_upsert_records (text records into an integrated index, as newline-delimited JSON) and pinecone_update_metadata change data; always confirm with the user first. pinecone_delete_vectors is installed switched off. Writes are visible to searches after a few seconds.

Costs: searches, queries and fetches use read units, writes use write units, embedding, reranking and assistant chat are billed per token or request on the user's Pinecone plan. Keep top_k small.

Errors: 401 "Invalid API key" or "Wrong API key" means the key is wrong, deleted or from another project. 403 means the key's role does not allow this (e.g. a viewer key writing). 404 is an unknown index, namespace or assistant; a DNS or connection error on a data tool usually means index_host was passed with .pinecone.io or https://. 400 on pinecone_search_records with text means the index is not an integrated index. 429 is a rate or plan limit.

Out of scope here: creating, configuring or deleting indexes, collections and backups, bulk imports, uploading or deleting assistant files, and project or key administration.