Pinecone N8n

Pinecone n8n Workflow Skill

<<invoke_this_skill>>

This skill helps you build n8n workflows with Pinecone nodes following best practices. It covers two Pinecone nodes:

• Pinecone Assistant (@pinecone-database/n8n-nodes-pinecone-assistant) — recommended for most use cases
• Pinecone Vector Store (@n8n/n8n-nodes-langchain.vectorStorePinecone) — for advanced control

Core rule: Always use the node’s built-in resources and operations. Never suggest using the HTTP node to call the Pinecone REST API directly.

Step 1: Understand the user’s scenario

Ask the user what they’re trying to do:

• Build a new workflow from scratch
• Configure or understand a specific Pinecone node
• Debug a workflow that isn’t working
• Review an existing workflow for best practices

Step 2: Node selection (for new workflows and configuration questions)

Always present the Pinecone Assistant node as the recommended choice first. Do NOT skip this step based on your own inference about which node fits better — even if the use case mentions specific triggers (Google Drive, webhooks, etc.) or file types (text, markdown, PDF), those details do not determine which node to use.

Only skip this step if:

• The user explicitly names a specific node (e.g. “I want to use the Vector Store node”, “help me set up pineconeAssistant”)
• The user is debugging or configuring an existing workflow that already has a specific Pinecone node in it

If the user has not named a node, always ask or recommend the Assistant node first. If the user said “use defaults” or you cannot ask, default to the Pinecone Assistant node and proceed with the Assistant path.

Ask the user which node they want to use, presenting these two options:

Pinecone Assistant (Recommended)

• Fully managed RAG — Pinecone handles chunking, embedding, and indexing automatically
• Built-in citations with file names and URLs
• Simpler setup: no embedding model or text splitter needed in n8n
• Great for: document Q&A, chat with files, knowledge base search

Pinecone Vector Store

• Full control over embedding model, chunking strategy, and metadata
• Works with any embedding model (OpenAI, Cohere, HuggingFace, etc.)
• Required when: you need custom embeddings, have an existing Pinecone index, need metadata filtering, or need fine-grained control over chunking

Pinecone Assistant Node — Best Practices and Workflow Generation

Node package names

• File operations (upload, list, delete): @pinecone-database/n8n-nodes-pinecone-assistant.pineconeAssistant
• Chat/retrieval as AI Agent tool: @pinecone-database/n8n-nodes-pinecone-assistant.pineconeAssistantTool

Prerequisites

• Create a Pinecone Assistant in the Pinecone Console at https://app.pinecone.io/organizations/-/projects/-/assistant (opens in a new tab) before running the workflow
• Set up a Pinecone credential in n8n with your API key

Workflow architecture

The standard pattern is a two-phase workflow:

Phase 1 — Ingestion (run once or on a schedule):

Manual Trigger → Set file URLs → Split Out → HTTP Request (download) → Pinecone Assistant (uploadFile)

Phase 2 — Chat:

Chat Trigger → AI Agent ← Pinecone Assistant Tool (connected as ai_tool)
                        ← OpenAI Chat Model (connected as ai_languageModel)

Key configuration rules

1. assistantData parameter: Always include BOTH name and host fields:

   jsonCopy

   {"name": "your-assistant-name", "host": "https://your-assistant-host.pinecone.io"}

   Find your assistant’s host in the Pinecone Console: open the assistant detail page and copy the host URL (format: https://<region>-data.<subdomain>.pinecone.io).
2. sourceTag: Always include in additionalFields:

   jsonCopy

   {"sourceTag": "n8n:n8n_nodes_pinecone_assistant:pinecone_n8n_skill"}

3. Connection type: The Assistant Tool connects to the AI Agent via the ai_tool connection — NOT main
4. externalFileId: Set this to the file URL expression so Pinecone stores it as a reference for citations
5. Credential: Use pineconeApi credential type for both node variants
6. File metadata on upload: Add key-value metadata via additionalFields.metadata.metadataValues — an array of {"key": "...", "value": "..."} objects. The externalFileId is automatically added to metadata; do not include it manually. Example:

   jsonCopy

Generating workflow JSON for the Assistant path

Build the workflow to match what the user actually describes — their triggers, models, data sources, and structure. Ask about anything structurally significant they haven’t mentioned. Only fall back to the defaults below when the user hasn’t specified a value:

• Assistant name: n8n-assistant (use n8n-assistant-1, n8n-assistant-2, etc. for multiples; must match an existing assistant in the Pinecone Console)
• File URLs: sample Pinecone release notes PDFs
• LLM model: gpt-5-mini
• System message: generic prompt about retrieving from the assistant with citations

The JSON below is a reference configuration showing correct parameter values, required fields, and connection types for each node. Use it as a guide for how to configure the nodes — not as a template to copy verbatim. Placeholders to substitute:

• [ASSISTANT_NAME] — assistant name
• [ASSISTANT_HOST] — assistant host URL from the Pinecone Console (e.g. https://your-assistant-host.pinecone.io)
• [USER_FILE_URLS_ARRAY] — JSON array of file URL strings, e.g. ["https://example.com/doc.pdf"]
• [USER_MODEL] — LLM model name, e.g. gpt-5-mini
• [USER_TOPIC] — short description of what the assistant knows, for the system message

Other Pinecone Assistant operations

Upload File vs Update File — choosing the right operation

updateFile is idempotent: it finds all files with the matching externalFileId, deletes them, then uploads the new version. If no file exists with that ID it behaves exactly like uploadFile. Use updateFile whenever a workflow may run more than once on the same source files.

Parameters for updateFile are identical to uploadFile: assistantData, externalFileId, inputDataFieldName, and all additionalFields including metadata, multimodalFile, and sourceTag.

{
  "parameters": {
    "resource": "file",
    "operation": "updateFile",
    "assistantData": "{\"name\":\"[ASSISTANT_NAME]\",\"host\":\"https://[ASSISTANT_HOST]\"}",
    "externalFileId": 

List Files

Use resource: "file", operation: "listFiles" to retrieve files, optionally filtered by metadata.

{
  "parameters": {
    "resource": "file",
    "operation": "listFiles",
    "assistantData": "{\"name\":\"[ASSISTANT_NAME]\",\"host\":\"https://[ASSISTANT_HOST]\"}",
    "additionalFields"

For complex filters, use advancedMetadataFilter (a JSON string) instead of metadataFilter — never both:

"additionalFields": {
  "advancedMetadataFilter": "{\"department\": {\"$in\": [\"legal\", \"finance\"]}}"
}

Get Context Snippets

Use resource: "contextSnippet", operation: "getContextSnippets" to retrieve relevant text or image chunks directly — useful when you need raw retrieved context as workflow data rather than a chat reply.

{
  "parameters": {
    "resource": "contextSnippet",
    "operation": "getContextSnippets",
    "assistantData": "{\"name\":\"[ASSISTANT_NAME]\",\"host\":\"https://[ASSISTANT_HOST]\"}",

The same metadataFilter / advancedMetadataFilter options work on getContextSnippets to scope retrieval to files matching specific metadata. The includeBinaryContent flag only applies when includeMultimodalContext is true.

Pinecone Vector Store Node — Best Practices and Workflow Generation

Node package name

@n8n/n8n-nodes-langchain.vectorStorePinecone

Prerequisites

• Create a Pinecone index in the Pinecone Console at https://app.pinecone.io/organizations/-/projects/-/indexes (opens in a new tab) with the correct name and dimensions before running the workflow
• Set up a Pinecone credential in n8n with your API key
• Set up an OpenAI credential in n8n

Workflow architecture

Two-phase workflow with two separate vectorStorePinecone node instances:

Phase 1 — Ingestion (run once or on a schedule):

Manual Trigger → Set file URLs → Split Out → HTTP Request (download)
  → Pinecone Vector Store (insert mode)
       ↑ Default Data Loader ← Recursive Character Text Splitter
       ↑ Embeddings OpenAI

Phase 2 — Chat:

Chat Trigger → AI Agent ← Pinecone Vector Store (retrieve-as-tool mode)
                               ↑ Embeddings OpenAI (same model as insert)
                        ← OpenAI Chat Model

Key configuration rules

1. Two node instances: Use one vectorStorePinecone in insert mode for ingestion and a separate one in retrieve-as-tool mode for chat
2. Embedding model consistency: The SAME embedding model and dimensions MUST be used in both insert and retrieve-as-tool nodes — they share a single Embeddings OpenAI node via the ai_embedding connection
3. Required companion nodes for the insert node:
   • @n8n/n8n-nodes-langchain.embeddingsOpenAi — set dimensions to match your index (1536 for text-embedding-3-small)
   • @n8n/n8n-nodes-langchain.textSplitterRecursiveCharacterTextSplitter — connects to the data loader
   • @n8n/n8n-nodes-langchain.documentDefaultDataLoader — set dataType: binary and textSplittingMode: custom
4. toolDescription: Write a clear, specific description on the retrieve-as-tool node — the AI Agent uses this to decide when to query it
5. topK: Use 20 as the default; increase for broader recall, decrease to reduce token usage
6. Metadata for citations: In the Default Data Loader, add external_file_url as a metadata field set to ={{ $json.urls }} so the AI Agent can cite sources
7. Credential: Use pineconeApi credential type

Chunking guidance

• 3000 chars / 500 overlap: Good default for long documents (articles, reports, documentation)
• 500–1000 chars / 100–200 overlap: Better for Q&A-style retrieval or short, dense content
• Reference: https://www.pinecone.io/learn/chunking-strategies/ (opens in a new tab)

Index setup instructions

When generating a workflow, always tell the user to:

1. Go to https://app.pinecone.io/organizations/-/projects/-/indexes (opens in a new tab) and create an index
2. Set the name to match the pineconeIndex value in the workflow
3. Select the embedding model that matches the n8n Embeddings node (text-embedding-3-small → 1536 dimensions)

Generating workflow JSON for the Vector Store path

Build the workflow to match what the user actually describes — their triggers, embedding model, chunking strategy, data sources, and structure. Ask about anything structurally significant they haven’t mentioned. Only fall back to the defaults below when the user hasn’t specified a value:

• Index name: n8n-index (use n8n-index-1, n8n-index-2, etc. for multiples; must match an existing Pinecone index)
• File URLs: sample Pinecone release notes PDFs
• LLM model: gpt-5-mini
• Embedding model: text-embedding-3-small (1536 dimensions)
• Chunk size: 3000 chars / 500 overlap
• System message / tool description: generic prompt about retrieving from the index with citations

The JSON below is a reference configuration showing correct parameter values, required fields, and connection types for each node. Use it as a guide for how to configure the nodes — not as a template to copy verbatim. Placeholders to substitute:

• [INDEX_NAME] — index name
• [USER_FILE_URLS_ARRAY] — JSON array of file URL strings, e.g. ["https://example.com/doc.pdf"]
• [USER_MODEL] — LLM model name, e.g. gpt-5-mini
• [USER_TOPIC] — short description of the data, for the AI Agent system message
• [USER_TOOL_DESCRIPTION] — description of what data the vector store contains, for the tool node

Debugging guidance

When a user reports a broken workflow, ask which node is failing and what error they see, then refer to these common issues:

Best practices review checklist

When reviewing an existing workflow:

1. No HTTP nodes calling Pinecone endpoints — replace any with the proper Pinecone node
2. Vector Store: same Embeddings node connected to both insert and retrieve-as-tool instances
3. Assistant: assistantData includes both name and host; sourceTag is present in additionalFields
4. toolDescription is specific and descriptive (not generic like “a vector store”)
5. Vector Store metadata: external_file_url stored in the Default Data Loader metadata for source citations
6. Assistant file metadata: files uploaded with meaningful key-value metadata when filtering will be needed later
7. Metadata filter: listFiles and getContextSnippets use a metadata filter (not both filter fields at once) when scoping to a subset of files
8. Multimodal upload: multimodalFile: true is set on uploadFile when PDFs contain images or charts — this is required to index visual content and is NOT the default
9. Multimodal retrieval: includeMultimodalContext: true and includeBinaryContent: true on getContextSnippets when the workflow needs image chunks alongside text
10. Chunk size is appropriate for the content type (see chunking guidance above)
11. Index exists with correct dimensions before the workflow runs