Pinecone

Pinecone is the vector database DocLD uses for embedding storage and vector search. It powers semantic search over your documents. DocLD uses Pinecone's integrated embedding model, so you don't call a separate embedding API—Pinecone handles embedding generation and storage in one place.

Integrated Embeddings

DocLD uses Pinecone's integrated embeddings with llama-text-embed-v2. You don't call a separate embedding API—Pinecone handles:

Embedding happens server-side when documents are chunked and when queries are sent. No embedding API calls from your app.

Namespace and Scoping

DocLD uses Pinecone namespaces to scope search. Knowledge bases filter which documents are queried—only documents in the selected knowledge base are searched. Namespace and metadata filters isolate data per tenant or project when needed.

Setup

Pinecone is configured when you set up DocLD:

1. Create Pinecone account — Sign up at pinecone.io
2. Create index — DocLD provisions or connects to a Pinecone index
3. Configure embedding model — DocLD uses llama-text-embed-v2 by default
4. Connect — DocLD uses your Pinecone API key for upserts and queries

You don't manage the index directly; DocLD handles upserts and queries as part of document processing and RAG chat.

Scaling

Pinecone is managed infrastructure: no vector index to host or scale yourself. As you add documents, Pinecone scales the index. DocLD handles batching and rate limits for upserts during batch processing.

Why Pinecone

Migration Notes

If you migrate from another vector store, DocLD would need to re-chunk, re-embed, and re-upsert documents to Pinecone. Plan for re-processing time during migration.

Related Concepts

Pinecone hosts the index for embeddings and vector search. Namespace and metadata enable scoping. Knowledge bases filter which documents are searched.