Resources · n8n integration

n8n Pinecone Vector Store nodeConfigure Pinecone Vector Store in n8n.

The n8n Pinecone Vector Store node is where your documents stop being files and start being searchable knowledge. It carries 2 head parameters and 5 operation modes, from bulk insert to answering an agent's question. Built for anyone wiring a retrieval layer into n8n.

Try the Pinecone Vector Store node on n8nSee the 2 parameters

Verified Trustpilot reviews · AI, automation & growth agency

Why automate

What does the n8n Pinecone Vector Store node actually do?

A vector store keeps mathematical representations of your text instead of the text itself, so a search returns what means the same thing rather than what spells the same way. This node is the bridge between an n8n workflow and a Pinecone index: it pushes documents in, pulls ranked documents out, updates an entry by its ID, or sits behind an AI node as the place answers get looked up.

Three shapes cover most of what people build with it. The first is ingestion: a workflow picks up files, splits them, and runs the node in insert mode against a Pinecone index. Nothing answers questions yet, you are just filling the store. The second is direct search: load mode takes a Prompt, ranks the stored documents by similarity, and hands back the top matches, with Limit deciding how many come down the wire. The third is agent-facing: in retrieve-as-tool mode the node becomes a tool, an add-on resource an AI model can call on its own when a question needs facts it does not have. That last one is how a support bot built on the AI Agent node answers from your own documentation instead of guessing.

Mode choice is the whole design decision here. retrieve hands the store to a chain or a tool as a vector store, which is the pattern when a Question and Answer Chain does the summarizing. retrieve-as-tool plugs the store straight into an agent's tool connector, where the Name and Description you write are what the model reads to decide whether to call it. A vague description is the single most common reason an agent ignores a perfectly good index.

The node handles vectors, not embeddings. Turning text into numbers happens in a separate sub-node, a helper node that never runs on its own and only works when plugged into a port of a root node. That is where OpenAI or Google Gemini embedding models come in, and their provider bills those calls under its own terms. Swapping that model later means re-inserting, since vectors written by one model are not comparable with vectors written by another.

Two limits worth knowing before you commit. This node exposes 2 head parameters and nothing else at the top level: everything past that lives inside the mode you picked, so a field you saw in a tutorial may simply belong to another mode. And it works on documents, not on Pinecone administration. Creating an index, deleting one, listing what exists in your account: none of that is here. For agent-side patterns the AI Agent Tool page covers the other half of the wiring, and the n8n review goes into where the platform fits overall.

Connect

What do you need before the node runs?

  1. 01

    Create the Pinecone credential

    Open the Credentials menu in n8n and add the Pinecone credential, which asks for a single API Key field. You create it once and every workflow on that instance can select it. Get the key from your Pinecone console: open the project, go to API Keys, and copy the value shown there into the credential field.

  2. 02

    Point the node at an index

    Pinecone Index is a resource locator, meaning you either pick the index from the list n8n loads with your credential or type its identifier yourself. Picking from the list is the safer route while you are learning, since a typed value that does not match an existing index fails at run time rather than at save time.

  3. 03

    Pick the mode before anything else

    Operation Mode decides which fields the panel shows you. Set it first. Choose insert while you are loading documents, then switch a second copy of the node to load or retrieve-as-tool for the reading side. Filling a panel and then changing the mode means re-entering the fields the new mode asks for.

Parameters

Which parameters do you set on the node?

The Pinecone Vector Store node has 2 parameters. For each one: the node as you configure it in n8n, what the parameter changes, and our field notes.

01

Operation Mode

mode

What you see in n8n

Notes & use cases

This selector decides what the node does with your Pinecone index, and every other field follows from it.

Key parameters

  • Operation Mode: load (Get Many) ranks stored documents against a Prompt, with Limit capping the results and Include Metadata deciding whether source details travel with them.
  • Operation Mode: insert (Insert Documents) writes documents in, update (Update Documents) rewrites one entry found by its ID.
  • Operation Mode: retrieve (Retrieve Documents) opens the store to a chain, retrieve-as-tool (Retrieve Documents As Tool) opens it to an agent under a Name and Description.
Use cases
a nightly insert run absorbs new policy documents, a chat workflow in load mode answers against them. Rerank Results reorders matches on the way out.
02

Pinecone Index

pineconeIndex

What you see in n8n

Notes & use cases

Every mode needs to know which index it is talking to, and this is the only field that is required no matter what you selected above.

Key parameters

  • Pinecone Index: required, and a resource locator, so it takes either a value chosen from the loaded list or one you enter yourself, including an expression such as {{ $json.index }} when the index name arrives from an earlier node.
Use cases
one index per client keeps a multi-tenant assistant from leaking one account's documents into another's answers, and driving the field from the incoming item lets a single workflow serve all of them instead of one workflow per index.
Need help

Need help automating Pinecone Vector Store with n8n?

A person reads every message.

FAQ

Pinecone Vector Store in n8n, answered

01Is the Pinecone Vector Store node included in n8n?
Yes. It ships inside the n8n AI package, so there is nothing to install and nothing extra to pay on the n8n side, on n8n Cloud and on a self-hosted instance alike. Self-hosting runs on Docker or npm under the Community Edition and its Sustainable Use license, and a workflow behaves the same either way. Two costs do sit outside n8n: Pinecone itself, under whatever terms your account has, and the provider of the embedding model you plug in, which bills its own API calls. n8n adds nothing on top of either.
02What do you need for the n8n Pinecone Vector Store node to work?
Three things. A Pinecone credential, created once in the Credentials menu and reusable across every workflow on the instance. An index, selected in the Pinecone Index field, which is required in all 5 operation modes. And an embeddings model, a sub-node that converts text into vectors and never runs on its own. Without that model the node has nothing to store or compare. Insert and read workflows should use the same embeddings model, because vectors produced by different models cannot be ranked against each other.
03What are the limits of this node?
It works on documents inside an index, not on the index itself. Creating, deleting or listing indexes in your Pinecone account is out of scope, and so is anything the node's 2 head parameters and 5 modes do not cover. Update works on one entry at a time, identified by its ID, so there is no bulk rewrite. The page describes version 1 of the node, the highest in the catalog; an older workflow may show an earlier version with fewer options. Namespace and metadata filtering live in the node options rather than in the main fields.
04What is the minimum you have to plug in to get started?
An embeddings model, and that is genuinely it for a first insert. Pick the embedding sub-node from the provider you already have a key for, connect it, set Operation Mode to insert, choose your index, and push a few documents through. Then build the reading half: a second node in load mode with a Prompt and a Limit tells you in one execution whether the store actually found the right passages. Only once that works is it worth handing the store to an agent as a tool, where a bad result is much harder to trace.
05n8n or Make for a retrieval workflow?
It depends on where your documents are allowed to live. n8n self-hosts on your own machine, so the text, the vectors and the model keys stay under your control, and its pricing does not count individual steps. Make is hosted only, with no self-hosting option, and bills per operation, which a retrieval pipeline running over many documents can reach quickly. Make's visual canvas is more guided; n8n asks more of you and gives more back on branching logic. Both connect to Pinecone, so the decision is about hosting, data control and cost shape.
Hack'celeration Lab

Get our weekly integration tips.

No spam. Unsubscribe anytime.

Same platform

Other docs you might need on this stack.

Browse all integrations