canonical: https://jentic.com/apis/pinecone.io/pinecone # Pinecone API Pinecone API provides a managed vector database for similarity search and AI applications. The API enables creating and managing vector indexes, upserting high-dimensional vectors with metadata, querying by vector similarity, and managing collections for index snapshots. It supports 7 cloud regions and handles index scaling, replication, and infrastructure management automatically across 15 endpoints. ## For AI agents Store, query, and manage high-dimensional vectors for similarity search, recommendation systems, and retrieval-augmented generation (RAG) applications. ## Scope Does not handle embedding generation, text processing, or model inference - use for vector storage and similarity search only. ## Capabilities - Upsert vectors with metadata into managed indexes for similarity search - Query indexes by vector similarity to find nearest neighbors - Create and configure vector indexes with custom dimensions and metrics - Manage collections as point-in-time snapshots of index data - Fetch specific vectors by ID for exact retrieval - Delete vectors by ID or metadata filter for index maintenance - Retrieve index statistics including vector counts and dimension info ## Use cases ### Retrieval-Augmented Generation (RAG) Store document embeddings in Pinecone and retrieve relevant context for LLM prompts. The /query endpoint accepts a vector and returns the most similar stored vectors with their metadata, enabling AI applications to ground responses in factual content from a knowledge base without fine-tuning. Example prompt: Upsert 100 document chunk embeddings with source metadata into an index, then query with a user question embedding to retrieve the top 5 most relevant chunks ### Semantic Search Engine Build semantic search by indexing content embeddings and querying with natural language. Pinecone returns results ranked by vector similarity rather than keyword matching, enabling search that understands meaning. The API handles index scaling automatically as the corpus grows to millions of vectors. Example prompt: Create a 1536-dimension index, upsert product description embeddings, and query with a natural language search embedding to find matching products ### Recommendation System Power recommendations by storing item embeddings and querying for similar items. When a user interacts with content, query Pinecone with that item's vector to find similar items the user might like. Metadata filters allow constraining results by category, price range, or availability. Example prompt: Query the index with a product embedding vector and filter results to items in the same category with price under $50 ### AI Agent Knowledge Retrieval AI agents use the Pinecone API through Jentic to store and retrieve knowledge vectors without managing infrastructure. Agents search for vector operations by intent, load the endpoint schema, and execute calls with Jentic handling API key authentication and region routing. Example prompt: Search Jentic for 'query vector similarity search', load the Pinecone /query endpoint schema, and find the top 10 similar documents for a query embedding ## Key endpoints | Method | Path | Description | | --- | --- | --- | | POST | /vectors/upsert | Insert or update vectors in an index | | POST | /query | Query an index with a vector for nearest neighbors | | GET | /vectors/fetch | Fetch vectors by ID | | POST | /vectors/delete | Delete vectors by ID or filter | | POST | /vectors/update | Update a vector's values or metadata | | POST | /describe_index_stats | Get index statistics | | POST | /databases | Create a new index | | GET | /databases | List all indexes | ## Key resources - **Indexes** — Create, configure, list, and delete vector indexes - **Vectors** — Upsert, query, fetch, update, and delete vectors - **Collections** — Create point-in-time snapshots of indexes for backup and migration - **Index Stats** — Retrieve index statistics including vector count and dimensions ## Why Jentic - **Setup:** Wiring the Pinecone API by hand means setting up its API-key header auth and resolving the controller host for the region your key belongs to before coding each upsert, query, and index call. Through Jentic you install once, import Pinecone from the API Directory, store the API key once, and your agent calls it. - **Permission scoping:** Pinecone takes vector and index details in the request body rather than a resource id in the path, so scoping is by operation: you limit the agent to the operations it needs, such as upsert and query for semantic search, and leave the rest out. You choose that set, so destructive ones like deleting vectors or an index are not included unless you add them. - **Credential handling:** Your Pinecone API key is stored once, encrypted, by your own Jentic One instance and injected at execution time, with the regional host resolved for you. It never enters the agent's prompt, logs, or context. - **Discovery method:** Agents search Jentic by intent such as 'store embeddings for semantic search' or 'query the nearest vectors', and Jentic returns the matching Pinecone operation with its input schema so the agent calls the right endpoint without navigating multiple regional environments. ## Related APIs - **Chroma API** — Chroma is an open-source embedding database that can run locally without cloud infrastructure - **Algolia Search API** — Algolia provides keyword search and faceted filtering that complements Pinecone's vector similarity search - **Apify API** — Apify scrapes and extracts web content that can be embedded and stored in Pinecone for RAG ## FAQ ### What authentication does the Pinecone API use? The Pinecone API uses an API key passed in the request header. Each key is scoped to a project and environment. Through Jentic, this key is stored encrypted in your Jentic One instance and injected automatically into requests. ### Can I use Pinecone for retrieval-augmented generation (RAG)? Yes. Upsert document chunk embeddings via /vectors/upsert with source metadata, then query with a user question embedding via /query to retrieve the most relevant chunks. The returned metadata provides the context for LLM grounding. ### What vector dimensions does Pinecone support? Pinecone supports vectors up to 20,000 dimensions. Common configurations use 1536 dimensions for OpenAI embeddings or 768 dimensions for sentence-transformers models. You specify the dimension when creating an index via POST /databases. ### How do I query for similar vectors through the Pinecone API using Jentic? Search Jentic for 'query vector similarity search', load the /query operation schema, and execute with your query vector, top_k count, and optional metadata filters. Jentic handles API key injection and region routing. ### What are the rate limits for the Pinecone API? Rate limits depend on your plan and pod type. The free tier allows 100 vectors per upsert call and limited queries per second. Paid plans scale based on the number of pods and replicas configured for each index. ### Can I filter query results by metadata? Yes. The /query endpoint accepts a filter parameter with metadata conditions. You can filter by exact match, range, or set membership on any metadata field attached to your vectors, combining similarity search with structured filtering. ### Can I limit what my agent is allowed to do with the Pinecone API? Yes. Because Jentic One is self-hosted by you, your own rules decide which Pinecone operations and credentials the agent may use. Pinecone passes vector and index details in the request body rather than a resource id in the path, so you scope access by operation: grant only the calls the agent needs, such as /vectors/upsert and /query for semantic search. Destructive operations like /vectors/delete or deleting an index stay off the list unless you explicitly add them.