Pinecon - Long-term Memory for AI

Pinecone는 고성능 벡터 검색 어플리케이션을 쉽게 만들 수 있게 해주는 관리형 클라우드 벡터 데이터베이스입니다. 사용하기 쉬운 API와 인프라에 대한 걱정 없이 초저지연 쿼리 처리와 라이브 인덱스 업데이트, 메타데이터 필터링 등 다양한 기능을 제공합니다.

Pinecone는 텍스트, 이미지, 제품 추천 등 다양한 분야에서 사용할 수 있으며, 벡터 검색 기능은 전통적인 키워드 기반 검색 방법과는 다르게, 벡터 임베딩을 이용해 데이터를 처리합니다. 이를 통해, 검색 쿼리와 가장 유사한 항목을 찾아내는데, 이 때 벡터 임베딩이 필요합니다. Pinecone는 밀집 임베딩과 희소 임베딩을 모두 지원합니다.

Pinecone를 이용하면, 텍스트 데이터를 변환하여 검색, 질문에 대한 답변을 생성, 이미지 유사도 검색, 제품 추천 등 다양한 어플리케이션을 쉽게 개발할 수 있습니다. 또한, 벡터 데이터베이스를 이용하면, 데이터 관리와 검색 쿼리 처리 등의 기능을 강화할 수 있습니다.

The Pinecone vector database makes it easy to build high-performance vector search applications. Developer-friendly, fully managed, and easily scalable without infrastructure hassles.

What is a Vector Database? | Pinecone

 

https://docs.pinecone.io/docs/overview

Overview

Pinecone makes it easy to build high-performance vector search applications. It’s a managed, cloud-native vector database with a simple API and no infrastructure hassles.

Pinecone has the following attributes:

• Fast: Get ultra-low query latency at any scale, even with billions of items.
• Fresh: Get live index updates when you add, edit, or delete data.
• Filtered: Combine vector search with metadata filters for more relevant, faster results.
• Fully managed: Get started, use, and scale with ease, while we keep things running smoothly and securely.

Get started using Pinecone.

Use cases

Pinecone is useful for a broad variety of applications. The following are some of the most common:

• Semantic text search: Convert text data into vector embeddings using an NLP transformer such as a sentence embedding model, then index and search through those vectors using Pinecone.
• Generative question-answering: Retrieve relevant contexts to queries from Pinecone and pass these to a generative model like OpenAI to generate an answer backed by real data sources.
• Hybrid search: Perform semantic and keyword search over your data in one query and combine the results for more relevant results.
• Image similarity search: Transform image data into vector embeddings and build an index with Pinecone. Then convert query images into vectors and retrieve similar images.
• Product recommendations: Generate product recommendations for ecommerce based on vectors representing users.

Want to see more and start with working example notebooks? See our example applications.

Key concepts

Vector search

Unlike traditional search methods that revolve around keywords, vector databases index and search through ML-generated representations of data, called vector embeddings, to find items most similar to the query.

Vector embeddings

Vector embeddings are sets of numbers that represent objects. They are generated by embedding models trained to capture the semantic similarity of objects in a given set. Pinecone supports two kinds of vector embeddings: dense embeddings and sparse embeddings.

You need to have vector embeddings to use Pinecone.

Vector database

A vector database indexes and stores vector embeddings for efficient management and fast retrieval. Unlike a standalone vector index, a vector database like Pinecone provides additional capabilities such as index management, data management, metadata storage and filtering, and horizontal scaling.

Learn more about vector databases.

Workflow

Follow these guides to set up your index:

1. Create an index
2. Connect to an index
3. Insert the data and vectors into the index

Once you have an index with data, follow these guides to start using your index:

• Query the data
  • Filter the data
• Fetch data
• Insert more data or update existing vectors
• Manage the index
• Manage data