Skip to main content

​
Overview

Pinecone is a fully managed vector database designed for high-performance similarity search. It provides ultra-low latency, high availability, and easy scaling without infrastructure management.

​
Setup

1

Create Pinecone Account

Sign up at Pinecone Console
2

Create an Index

Create a new index with:
  • Index name (e.g., flowise-docs)
  • Dimensions (must match your embedding model, e.g., 1536 for OpenAI)
  • Metric (cosine, euclidean, or dotproduct)
3

Get API Key

Copy your API key from the Pinecone console
4

Add to Flowise

Create a Pinecone API credential in Flowise with your API key

​
Configuration

​
Required Parameters

​
credential
credential
required
Pinecone API key credential
​
pineconeIndex
string
required
Name of your Pinecone index (must be created beforehand)
​
embeddings
Embeddings
required
Embedding model to convert text to vectors (e.g., OpenAI Embeddings)

​
Optional Parameters

​
document
Document[]
Documents to upsert into the vector store
​
recordManager
RecordManager
Track records to prevent duplication during indexing
​
pineconeNamespace
string
Namespace for organizing vectors within an index. Useful for multi-tenancy:
pineconeNamespace: "user-123"
pineconeNamespace: "production"
​
pineconeTextKey
string
default:"text"
Metadata key for storing the document text content
​
pineconeMetadataFilter
json
Filter results by metadata during queries:
{
  "category": "documentation",
  "year": { "$gte": 2023 }
}
​
topK
number
default:4
Number of most similar results to return
​
fileUpload
boolean
Allow file uploads in chat, automatically scoped to chat session

​
MMR (Maximal Marginal Relevance)

​
searchType
string
default:"similarity"
Search algorithm:
  • similarity - Standard cosine similarity
  • mmr - Maximize diversity while maintaining relevance
​
fetchK
number
default:20
Number of documents to fetch before MMR reranking (only when using MMR)
​
lambda
number
MMR diversity factor (0 = max diversity, 1 = max relevance)

​
Usage Examples

​
Basic Setup

// Create index in Pinecone console first
// Name: flowise-kb
// Dimensions: 1536 (for OpenAI embeddings)
// Metric: cosine

// In Flowise node:
Pinecone Index: flowise-kb
Embeddings: OpenAI Embeddings
Top K: 4

​
With Namespaces

// Multi-tenant setup
Pinecone Index: shared-index
Pinecone Namespace: customer-abc
Embeddings: OpenAI Embeddings

// Different namespace for another customer
Pinecone Namespace: customer-xyz

​
Metadata Filtering

// Only search recent documentation
{
  "pineconeMetadataFilter": {
    "source": "docs",
    "updated_at": { "$gte": "2024-01-01" }
  }
}

​
With MMR for Diversity

// Get diverse results
Search Type: mmr
Fetch K: 20
Lambda: 0.5
Top K: 5

​
With Record Manager

// Prevent duplicate indexing
Document: PDF Loader output
Embeddings: OpenAI Embeddings
Record Manager: Postgres Record Manager
Pinecone Index: documents

// Record manager tracks what's been indexed
// Updates only changed documents

​
Metadata Filter Syntax

Pinecone supports rich metadata filtering: 
// Equality
{ "category": "tutorial" }

// Comparison operators
{
  "price": { "$gte": 10, "$lte": 50 },
  "rating": { "$gt": 4.0 }
}

// Array membership
{ "tags": { "$in": ["AI", "ML"] } }

// Logical operators
{
  "$and": [
    { "category": "docs" },
    { "year": { "$gte": 2023 } }
  ]
}

{
  "$or": [
    { "priority": "high" },
    { "urgent": true }
  ]
}

​
Best Practices

Index Design

  • Create one index per use case
  • Use namespaces for multi-tenancy
  • Match dimensions to embedding model
  • Choose appropriate distance metric

Performance

  • Use metadata filtering to reduce search space
  • Batch upserts for efficiency
  • Monitor query latency
  • Use appropriate pod type

Cost Optimization

  • Delete old vectors periodically
  • Use namespaces instead of multiple indexes
  • Right-size your pod type
  • Monitor storage usage

Data Management

  • Use record manager to avoid duplicates
  • Set up backup/export strategy
  • Tag vectors with metadata
  • Implement soft deletes if needed

​
Index Configuration Guide

​
Choosing Dimensions

Dimensions must match your embedding model:
Embedding ModelDimensions
OpenAI text-embedding-ada-0021536
OpenAI text-embedding-3-small512 or 1536
OpenAI text-embedding-3-large256, 1024, or 3072
Cohere embed-english-v3.01024
Cohere embed-multilingual-v3.01024

​
Distance Metrics

  • Cosine - Best for most use cases, normalized similarity
  • Euclidean - Absolute distance, good for specific domains
  • Dot Product - Fast, assumes normalized vectors

​
Pod Types

  • s1 - Storage-optimized (large datasets, lower cost)
  • p1 - Performance-optimized (lowest latency)
  • p2 - High-performance (very large scale)

​
Outputs

The Pinecone node provides two outputs:
​
retriever
VectorStoreRetriever
Use as a retriever in chains and agents. Automatically handles similarity search.
​
vectorStore
PineconeVectorStore
Direct vector store access for custom operations like adding/deleting vectors.

​
Common Issues

Error: “Vector dimension does not match index dimension”Solution:
  • Ensure embedding model dimensions match index
  • Create new index with correct dimensions
  • Check embedding configuration
Error: “Index ‘name’ not found”Solution:
  • Verify index exists in Pinecone console
  • Check index name spelling (case-sensitive)
  • Ensure index is in ‘Ready’ state
  • Wait for index initialization to complete
Empty results despite having dataSolution:
  • Check namespace spelling
  • Verify data was inserted into correct namespace
  • Try querying without namespace filter
  • Use Pinecone console to inspect data
Results ignore metadata filterSolution:
  • Ensure metadata was included during upsert
  • Check JSON syntax in filter
  • Verify metadata key names match
  • Test with simpler filter first

​
Monitoring & Maintenance

In the Pinecone console, monitor:
  • Vector count - Track total vectors per namespace
  • Query latency - p50, p95, p99 percentiles
  • Storage usage - Plan capacity
  • Request rate - API usage patterns

​
Scaling Considerations

Pinecone automatically scales, but consider:
  • Pod count - Increase for more QPS
  • Pod type - Upgrade for better performance
  • Replicas - Add for high availability
  • Namespaces - Use for logical separation

Build docs developers (and LLMs) love

Get started for freeTalk to us