pinecone

Pinecone RAG Sample

This sample demonstrates how to use Pinecone as a vector database for RAG (Retrieval Augmented Generation) workflows with Genkit.

Prerequisites

1. Pinecone Account: Sign up at pinecone.io
2. Pinecone API Key: Get from the Pinecone console
3. Google GenAI API Key: For embeddings and LLM

Environment Variables

Variable	Required	Default	Description
GEMINI_API_KEY	Yes	-	Google GenAI API key
PINECONE_API_KEY	Yes	-	Pinecone API key
PINECONE_INDEX_NAME	No	genkit-films	Name of the Pinecone index
PINECONE_CLOUD	No	aws	Cloud provider (aws, gcp, azure)
PINECONE_REGION	No	us-east-1	Region for the index

Running the Sample

Using .env file (Recommended)

1. Copy the example environment file:

   cp .env.example .env

2. Edit .env and set your API keys:

   GEMINI_API_KEY=your-gemini-api-key
   PINECONE_API_KEY=your-pinecone-api-key
   

3. Run the sample:

   ./run.sh

Using Environment Variables

# Set required environment variables
export GEMINI_API_KEY="your-gemini-api-key"
export PINECONE_API_KEY="your-pinecone-api-key"

# Optional: Custom index name
# export PINECONE_INDEX_NAME="my-custom-index"

# Run the sample
./run.sh

Or with Maven directly:

mvn exec:java -Dexec.mainClass="com.google.genkit.samples.pinecone.PineconeRAGSample"

Index Setup

The sample is configured to NOT auto-create indexes by default. You have two options:

Option 1: Create Index in Pinecone Console

1. Go to Pinecone Console
2. Create a new index with:
   • Name: genkit-films (or your custom name)
   • Dimensions: 768
   • Metric: cosine
   • Cloud: Your preferred cloud provider

Option 2: Enable Auto-Creation

Modify the sample code to set createIndexIfNotExists(true):

.addIndex(PineconeIndexConfig.builder()
    .indexName(indexName)
    .createIndexIfNotExists(true)  // Enable auto-creation
    // ...
    .build())

Available Flows

indexDocuments

Indexes sample documents about famous films into Pinecone.

curl -X POST http://localhost:4000/api/flows/indexDocuments \
  -H 'Content-Type: application/json' \
  -d '{}'

retrieveDocuments

Retrieves documents matching a semantic query.

curl -X POST http://localhost:4000/api/flows/retrieveDocuments \
  -H 'Content-Type: application/json' \
  -d '{"data": "sci-fi movies"}'

ragQuery

Answers questions using RAG with retrieved context.

curl -X POST http://localhost:4000/api/flows/ragQuery \
  -H 'Content-Type: application/json' \
  -d '{"data": "What Christopher Nolan films are mentioned?"}'

Example Workflow

1. Create an index in Pinecone (see Index Setup above)
2. Start the sample application
3. Index the sample documents:

   curl -X POST http://localhost:4000/api/flows/indexDocuments -H 'Content-Type: application/json' -d '{}'

4. Query for relevant documents:

   curl -X POST http://localhost:4000/api/flows/retrieveDocuments -H 'Content-Type: application/json' -d '{"data": "movies about dreams"}'

5. Ask questions using RAG:

   curl -X POST http://localhost:4000/api/flows/ragQuery -H 'Content-Type: application/json' -d '{"data": "Which films were directed by Christopher Nolan and what are they about?"}'

Configuration

The sample uses:

• Embedder: googleai/text-embedding-004 (768 dimensions)
• LLM: googleai/gemini-2.0-flash
• Metric: Cosine similarity
• Index Type: Serverless (AWS us-east-1)

You can modify these settings in PineconeRAGSample.java.

Using Namespaces

Pinecone supports namespaces for multi-tenant applications. To use namespaces:

.addIndex(PineconeIndexConfig.builder()
    .indexName("my-index")
    .namespace("production")  // Add namespace
    .embedderName("googleai/text-embedding-004")
    .build())

The retriever/indexer will be available at /retriever/pinecone/my-index/production.

Troubleshooting

Index not found

Index genkit-films does not exist and createIndexIfNotExists is false

Either create the index in the Pinecone console or enable auto-creation.

Dimension mismatch

Vector dimension does not match index dimension

Ensure your index was created with 768 dimensions to match text-embedding-004.

Rate limiting

Pinecone has rate limits on operations. For high-volume applications, implement retry logic with exponential backoff.

Cost Considerations

Pinecone charges based on:

• Number of vectors stored
• Number of queries
• Pod type (for pod-based indexes)

For development, serverless indexes on the free tier are recommended.