Skip to main content

Pinecone

The Pinecone package enables you to integrate Pinecone's vector database into your RPA automation workflows. Pinecone is a fully-managed, developer-friendly vector database designed for AI applications, enabling semantic search, recommendation systems, and Retrieval-Augmented Generation (RAG) workflows.

Features​

  • Create and manage vector database indexes
  • Insert and update high-dimensional vector embeddings
  • Perform similarity searches with filtering capabilities
  • Fetch vectors by ID or query by similarity
  • Delete vectors individually, by filter, or in bulk
  • Get detailed statistics about index usage and namespaces
  • Support for metadata filtering and namespaces for data organization

Prerequisites​

Before using the Pinecone package, you need to:

  1. Create a Pinecone account at pinecone.io
  2. Generate an API key from the Pinecone console at https://app.pinecone.io/organizations/-/projects/-/keys
  3. Store your API key as a credential in Robomotion
  4. Note your Pinecone environment (e.g., us-east-1-aws, us-west1-gcp)

Common Use Cases​

  • Retrieval-Augmented Generation (RAG) - Store document embeddings for AI chatbots and question-answering systems
  • Semantic Search - Build intelligent search engines that understand meaning, not just keywords
  • Recommendation Systems - Find similar products, content, or users based on vector similarity
  • Document Similarity - Compare and cluster documents based on their semantic content
  • Image Search - Store and search image embeddings for visual similarity
  • Anomaly Detection - Identify outliers by finding vectors that are dissimilar to the norm
  • Content Deduplication - Detect duplicate or near-duplicate content using similarity search

Getting Started​

The typical workflow involves:

  1. Connect - Establish connection with your Pinecone API key
  2. Create Index - Set up a new index with appropriate dimensions and metrics
  3. Upsert - Insert vector embeddings with metadata
  4. Query - Search for similar vectors
  5. Fetch - Retrieve specific vectors by ID
  6. Delete - Remove vectors when no longer needed

Alternatively, you can provide the API key directly to each node without using the Connect node.

Understanding Vector Databases​

Vector databases like Pinecone store high-dimensional vectors (arrays of numbers) that represent data in a mathematical space. When you use AI models like OpenAI's text-embedding-ada-002 or other embedding models, they convert text, images, or other data into vectors. These vectors capture semantic meaning, allowing you to:

  • Find similar items by calculating distance between vectors
  • Filter results using metadata (e.g., category, date, author)
  • Scale to billions of vectors with fast query performance

Index Configuration​

When creating an index, you need to specify:

  • Dimension - Must match your embedding model's output (e.g., 1536 for OpenAI ada-002)
  • Metric - Distance function for similarity:
    • cosine - Cosine similarity (most common, good for normalized vectors)
    • euclidean - Euclidean distance (good for absolute differences)
    • dotproduct - Dot product (good for speed when vectors are normalized)
  • Pod Type - Performance tier (p1, p2, s1)
  • Pods & Replicas - Scale and availability settings

Namespaces​

Namespaces allow you to partition vectors within a single index, useful for:

  • Multi-tenant applications (one namespace per user/organization)
  • Separating development, staging, and production data
  • Organizing vectors by category or data type