Does LanceDB send my documents to a cloud service?

No. LanceDB is an embedded, on-disk vector store and embeddings are produced locally with Ollama, so document content never leaves your machine. Pinecone is a managed cloud service, so your vectors live in Pinecone's infrastructure.

Do I have to generate embeddings myself?

With LanceDB you build the index ahead of time using local Ollama models via the seed command. With Pinecone's integrated inference you pass text and Pinecone generates the embeddings for upsert and search, so you don't manage an embedding model yourself.

LanceDB vs Pinecone

LanceDB and Pinecone are both vector databases with MCP servers, but they represent the embedded-and-local versus managed-and-cloud poles of vector search. The LanceDB server gives an agent agentic RAG over a local LanceDB index built on a two-level model — a catalog of document-level summaries and a chunk store of passages — with three focused tools: search the catalog for relevant documents, find chunks within a specific document, and find chunks across all documents. Because LanceDB is an embedded, on-disk store and embeddings are produced locally with Ollama, document content never leaves your machine, which suits private knowledge bases and offline RAG. Pinecone's official developer server works against your managed Pinecone vector database and its docs: list and describe indexes, read stats and namespaces, create an index backed by an integrated embedding model, upsert and search records using integrated inference (pass text, Pinecone embeds it), rerank, run a cascading search across multiple indexes with deduplication, and search the Pinecone docs. So one is private and embedded, the other managed and feature-rich. Here is how they compare.

How they compare

Dimension	LanceDB	Pinecone
Architecture	Embedded and on-disk: a local LanceDB index queried by the agent; document content never leaves your machine.	Managed cloud service: indexes hosted by Pinecone, reached over api.pinecone.io with serverless scaling.
Embeddings	Produced locally with Ollama (the README pulls snowflake-arctic-embed2 and llama3.1:8b); you build the index ahead of time with a seed command.	Integrated inference: you pass text and Pinecone handles the embeddings when you create an index for a model and upsert or search records.
Retrieval features	A two-level catalog/chunk model with three tools — catalog search, per-document chunk search, and all-document chunk search — to locate a source then drill into passages.	Search records, reranking, and a cascading search across multiple indexes with deduplication and reranking, plus index stats and namespace inspection.
Deployment and auth	Published to npm, run locally over stdio via npx (lance-mcp) pointed at a local index directory; needs Ollama running with the embedding/summarization models.	@pinecone-database/mcp run locally over stdio via npx with a PINECONE_API_KEY; a separate Assistant MCP server offers a managed remote context-retrieval endpoint with a bearer key.
Best-fit task	Private, offline RAG over a fixed document set where data must stay on the machine.	Production retrieval at scale where you want a managed service, integrated embeddings, reranking, and multi-index search.

Verdict

Pick by where your vectors must live and how much you want managed for you. Choose LanceDB when privacy and offline operation matter: it runs an embedded, on-disk index with local Ollama embeddings, so document content never leaves your machine, and its catalog/chunk model gives an agent a clean find-the-source-then-drill-in flow over a fixed corpus. Choose Pinecone when you want a managed, scalable service with batteries included — integrated embeddings so you pass text not vectors, reranking, namespaces, and cascading multi-index search — plus a separate Assistant endpoint for managed context retrieval. LanceDB is private and embedded; Pinecone is managed and feature-rich. The decision usually follows your data-residency and scale requirements.

FAQ

Does LanceDB send my documents to a cloud service?: No. LanceDB is an embedded, on-disk vector store and embeddings are produced locally with Ollama, so document content never leaves your machine. Pinecone is a managed cloud service, so your vectors live in Pinecone's infrastructure.
Do I have to generate embeddings myself?: With LanceDB you build the index ahead of time using local Ollama models via the seed command. With Pinecone's integrated inference you pass text and Pinecone generates the embeddings for upsert and search, so you don't manage an embedding model yourself.