Resources · n8n integration

n8n Vector Store Retriever nodeConfigure Vector Store Retriever in n8n.

The n8n Vector Store Retriever node turns a vector store into a searchable source for an AI chain. It exposes one parameter, Limit, and one required input port for the store itself. Its output feeds question answering chains, so answers stay anchored in documents.

Try the Vector Store Retriever node on n8nSee the 1 parameters

Verified Trustpilot reviews · AI, automation & growth agency

Why automate

What does the n8n Vector Store Retriever node actually do?

A vector store keeps mathematical representations of your documents, and a retriever is the piece that reads from it. This node sits between the two: you branch a vector store onto its Vector Store port, it queries that store with whatever question arrives, and it hands back the closest matching documents. It never runs on its own, because it is a sub-node, a helper node that only works once a root node pulls on it.

Start with a support knowledge base. Product PDFs get split, turned into embeddings (numeric fingerprints of each chunk of text) and stored. A Question and Answer Chain asks the retriever for the passages closest to the incoming question, and the chain writes the reply from those passages only. Nothing else changes in the workflow, and the Limit parameter decides how much context the chain reads.

Second scenario, an internal research assistant. A workflow drops meeting notes into a store every week. When somebody queries it, the retriever returns the handful of notes that match, and an AI Agent further down turns them into a short brief. Because the retriever hands back documents rather than prose, the same output works for a summary, a tag, or a routing decision.

Third, a compliance check. A chain compares an incoming request against stored policy text. The retriever pulls the closest clauses, the model quotes them, and a reviewer reads the source alongside the answer. Model calls go through whichever chat model sub-node the chain uses, from OpenAI or Anthropic, and that provider bills its own API calls under its own terms. n8n adds nothing on top.

When to reach for something else: if the root node you are building already talks to a vector store directly, this node adds a hop for nothing. It earns its place when a chain or another retriever expects a retriever on its input, which is exactly what the MultiQuery Retriever and the Contextual Compression Retriever do. Both wrap this node rather than replace it.

Known limits are worth stating plainly. The node reads, it never writes, so loading documents stays the job of the vector store node itself. It carries a single parameter at version 1, which means tuning happens in the store and in the chain around it, not here. For a broader look at where this fits in the platform, the n8n review covers the AI side in context.

Connections

What do you plug into this node?

n8n draws AI workflows as a root node with sub-nodes hanging off its ports, one type of sub-node per port. This node is a sub-node itself: it takes a store on one side and offers a retriever on the other.

Ports (what plugs in)

  • Vector Storeai_vectorStorerequired

Output (what it plugs into)

  • Retrieverai_retriever
01Vector StoreThis port receives the store the node reads from. Drag a vector store sub-node onto it and the retriever queries that store every time a root node calls it.

This port receives the store the node reads from. Drag a vector store sub-node onto it and the retriever queries that store every time a root node calls it.

Key parameters

  • Required: leave this port empty and the node blocks the run, because there is nothing to search.
Use cases
a knowledge base of product documentation, or a store filled weekly with fresh notes, becomes readable by any chain without duplicating the search logic.
02RetrieverThe output side. This node presents itself as a retriever, and root nodes that accept one pull the documents from it when they need context.

The output side. This node presents itself as a retriever, and root nodes that accept one pull the documents from it when they need context.

Key parameters

  • Optional: nothing runs until a root node connects here, since a sub-node never executes alone.
  • Question and Answer Chain: the shortest path, it asks a question and writes an answer from what comes back.
  • MultiQuery Retriever: rewrites one question into several before searching, then merges the results.
  • Contextual Compression Retriever: trims the returned documents down to the parts that matter for the question.
Use cases
the same store serves a simple question and answer flow today and a compression layer later, with only the root node swapped.
Parameters

Which parameter is worth your attention?

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

01

Limit

topK

What you see in n8n

Notes & use cases

Limit sets the maximum number of results the node returns on each search. The store ranks matches by closeness, and this number decides how far down that ranking the node reads before it stops.

Key parameters

  • Limit: a number, the maximum number of results to return. A value coming from an earlier node goes in as an expression, the n8n syntax for reading a field, written {{ $json.limit }}.
Use cases
a short factual question rarely needs more than a handful of passages, while a summary of a long policy reads better with more. Raising it feeds more text to the model downstream, so the trade sits between coverage and how much the chain has to read.
Need help

Need help automating Vector Store Retriever with n8n?

A person reads every message.

FAQ

Vector Store Retriever and n8n, the usual questions

01Is the Vector Store Retriever node free in n8n?
Yes. The node ships inside the n8n AI package, so it is there on n8n Cloud, the hosted offer run by n8n, and on a self-hosted instance under the Community Edition and its Sustainable Use license. There is nothing to install and n8n charges nothing extra for it. A workflow behaves the same way in both places. What can cost money sits elsewhere: the provider behind the chat model used by the chain around this node bills its own API calls under its own terms, and the vector store you branch onto the Vector Store port may have its own hosting arrangement.
02What do you need to authenticate to use it?
Nothing on this node. It has no credential and no Authentication selector, which is why the page has no connection section at all. The account details live one step away, on the sub-node branched onto the Vector Store port: that store carries whatever credential its own provider requires. Same story for the chat model a chain uses further along. So the setup work here is wiring, not account configuration. Drag the store onto the port, connect the output to a root node, and the node is ready to run.
03What are the limits of the n8n Vector Store Retriever node?
It reads, it does not write. Loading, splitting or inserting documents belongs to the vector store node and the loaders around it, not here. The node is at version 1 and carries a single parameter, Limit, so there is no filtering, no re-ranking and no scoring option to set on it. An older workflow may show an earlier version with fewer options. It also never runs by itself: as a sub-node it waits for a root node to call it, and a test click on the node alone will not produce documents.
04What is the minimum you have to connect, and where should you start?
Two connections. A vector store on the required Vector Store port, and a root node on the retriever output. For a first build, the Question and Answer Chain is the easiest root node to pair with: it takes a question, calls the retriever, and writes an answer from the documents that come back. Once that works, the MultiQuery Retriever and the Contextual Compression Retriever are the two natural upgrades, one widening the search, the other trimming what comes back. Neither replaces this node, both sit around it.
05n8n or Make for a retrieval workflow?
It depends on where the data has to live. n8n runs self-hosted with Docker or npm, or on n8n Cloud, so a workflow that reads internal documents can stay on your own infrastructure. Make is hosted only, with no self-hosting option, and it bills per operation, which changes the arithmetic on a flow that fires often. Make also has a visual logic many teams find quicker to read at a glance. Both build automations that work. The choice is about hosting, control over the data, and the cost model rather than one being better than the other.
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