Skip to main content

Pinecone Integration

Pinecone is the vector database powering the RAG (Retrieval-Augmented Generation) system for AI email generation.

Setup​

Two Pinecone indexes are required:

IndexDimensionsEmbedding Model
Primary1536text-embedding-3-small (OpenAI)
Large3072text-embedding-3-large (OpenAI)

Both indexes are queried and updated simultaneously.

Configuration​

PINECONE_API_KEY=your-api-key
PINECONE_INDEX=your-index-name           # 1536-dim
PINECONE_INDEX_LARGE=your-large-index    # 3072-dim

Vector Metadata​

Each vector in Pinecone stores metadata used for filtering. The metadata structure varies by content type:

HTML Code Examples (type: 'html')​

{
  "type": "html",
  "description": "Product showcase with tabbed navigation",
  "technique": "tabs",
  "complexity": "intermediate",
  "htmlType": "complete",
  "emailPurpose": "ecommerce",
  "exampleType": "positive",
  "keyFeatures": ["lightswitch", "mobileResponsive"],
  "bestPracticeTags": ["tableStructure", "msoConditionals"],
  "html": "<!DOCTYPE html>...",
  "notes": "",
  "submittedAt": "2025-01-15T..."
}

AMP Code Examples (type: 'amp')​

{
  "type": "amp",
  "description": "AMP tabbed product showcase",
  "technique": "tabs",
  "complexity": "intermediate",
  "htmlType": "complete",
  "html": "<!doctype html><html ⚡4email>...",
  "notes": "",
  "submittedAt": "2025-01-15T..."
}

Blog Articles (type: 'blog')​

{
  "type": "blog",
  "contentFocus": "kinetic",
  "blogTitle": "Building Accessible Tab Interfaces in Email",
  "blogContent": "Full article text...",
  "blogTopic": "kinetic-techniques",
  "learningLevel": "intermediate",
  "techniquesCovered": ["tabs", "accessibility"],
  "keyTakeaways": "Summary of key learnings...",
  "submittedAt": "2025-01-15T..."
}

The contentFocus field controls which build type receives the blog during generation:

ValueIncluded In
kineticKinetic builds only
ampAMP builds only
generalBoth kinetic and AMP builds

Blogs without contentFocus default to general.

Query Patterns​

Technique-Filtered Query​

# Pinecone query with metadata filter
index.query(
  vector=embedding,
  top_k=10,
  filter={"technique": {"$eq": "tabs"}},
  include_metadata=True
)

Broad Unfiltered Query​

index.query(
  vector=embedding,
  top_k=15,
  include_metadata=True
)

Content Management​

Admin endpoints manage Pinecone content:

EndpointAction
POST /api/admin/submit-contentUpload new content (embeds + stores in both indexes)
POST /api/admin/update-contentRe-embed and update existing content
POST /api/admin/delete-contentRemove from both indexes
GET /api/admin/list-contentBrowse content with filters

See RAG Overview for the full retrieval pipeline.