pinecone:n8n

Use the Pinecone Assistant node for managed RAG or the Pinecone Vector Store node for custom control. The standard pattern is a two-phase workflow: an ingestion phase that downloads and uploads files, and a chat phase connecting an AI Agent to the Pinecone tool node.

Use the Assistant node for fully managed RAG with automatic chunking, embedding, and citations. Use the Vector Store node when you need custom embedding models, an existing Pinecone index, metadata filtering, or fine-grained chunking control.

The assistantData name parameter does not match an existing assistant. Create the assistant in the Pinecone Console first, then set the name field exactly and include the assistant host URL in the assistantData parameter.

The toolDescription on the retrieve-as-tool node is too vague for the agent to select it. Write a specific description stating what data the store contains and what questions it can answer.

The embedding model dimensions must match the Pinecone index dimensions. For text-embedding-3-small, set the Embeddings OpenAI node to 1536 dimensions and create the index with the same value.

Yes, listFiles and getContextSnippets support metadataFilter with key-value pairs for equality matching, or advancedMetadataFilter as a JSON string for operators like $in and $or. You cannot set both filter fields at once.