Resources · n8n integration

n8n Embeddings OpenAI nodeConfigure Embeddings OpenAI in n8n.

Every vector store in n8n needs something to turn text into numbers first. The n8n Embeddings OpenAI node is that piece: a sub-node with 7 parameters, no main input of its own, that plugs into the embeddings port of a root node. Built for anyone wiring up a search index or a RAG chatbot.

Try the Embeddings OpenAI node on n8nSee the 7 parameters

Verified Trustpilot reviews · AI, automation & growth agency

Why automate

What does the n8n Embeddings OpenAI node actually do?

It converts text into embeddings, the vectors that let a database compare meaning instead of matching words. It never runs on its own. You attach it to a root node that needs embeddings, that root node feeds it the text, and it hands vectors back. Nothing appears on the canvas after it, because it has no main output of its own.

The clearest scenario is a document search. A Default Data Loader reads a file, a Token Splitter cuts it into chunks, and a Qdrant Vector Store stores them. The store itself has no idea how to vectorize anything, so Embeddings OpenAI sits underneath it and does that step. Templates on n8n.io build exactly this shape, workflow 2846 among them.

The second scenario is answering questions over that index. An AI Agent gets a Vector Store Question Answer Tool, the tool queries the same store, and the same embeddings node runs again so the question is vectorized the same way the documents were. If you swap the model or the dimensions value between indexing and querying, the two sets of vectors stop lining up and the answers go strange.

The third one is smaller and often overlooked: a newsletter or digest that scores incoming items for similarity before anything is written. Workflow 3986 pairs an agent with a Simple Vector Store, RSS content, and Gmail at the end.

When would you use something else? If your goal is a chat completion, a transcription or an image, this is the wrong node and the OpenAI node is the one you want. Embeddings OpenAI does one job and exposes no operations at all, only options.

The known limits are worth reading before you commit. The node has a single version in the catalog, version 1, and its whole surface is the 7 parameters below. There is no retry setting, no chunking, no text cleanup beyond Strip New Lines. Chunk size belongs to the splitter node, not here. And the model provider bills its own API calls under its own terms, while n8n adds nothing on top. If you want a longer look at the platform before going further, the n8n review and the n8n training pages cover it.

Connect

What do you connect before the node will run?

  1. 01

    Open the Credentials menu

    The node reads an OpenAI credential, and that credential lives in n8n rather than in the node. Create it once from the Credentials menu and every workflow on the instance can reuse it, including the chat model nodes that sit next to this one in a RAG setup. Nothing is stored inside the workflow itself, so exporting the workflow never carries the key with it.

  2. 02

    Paste the API key

    The credential page for OpenAI is the same one the other OpenAI nodes use, so if you already built one for a chat model, select it and you are done. A fresh key is created on the OpenAI side, then pasted into the credential form in n8n. Keep in mind that the provider charges its own API calls according to its own terms; n8n does not resell or mark up anything.

  3. 03

    Attach the node to a root node

    Drag the node onto the canvas and connect it to the embeddings port of a vector store or a similar root node. Until that link exists, the node has nowhere to send its vectors and the workflow will not execute it. There is no main input line to draw, only the port connection underneath the root node.

Connections

Where does this sub-node plug in?

n8n splits AI work between a root node, which receives the workflow items, and sub-nodes attached to its ports, one type of sub-node per port. This node is a sub-node and has no main input at all.

Output (what it plugs into)

  • Embeddingsai_embedding
01EmbeddingsThe node exposes a single output port of type ai_embedding. A root node that indexes or searches documents opens an embeddings slot underneath itself, and this node fills it.

The node exposes a single output port of type ai_embedding. A root node that indexes or searches documents opens an embeddings slot underneath itself, and this node fills it.

Key parameters

  • Required: the node never runs alone, so the port has to be connected to a root node before anything executes.
Use cases
the published templates show both sides of that link. A Qdrant Vector Store carries the indexing and the querying in workflows 2846 and 2845, a Simple Vector Store does the same in memory in workflow 3986, and a Vector Store Question Answer Tool hangs off an agent when the index is read rather than built. A support bot that reads a Google Drive folder uses this node in both phases.
Parameters

Which parameter do you change, and when?

The Embeddings OpenAI node has 7 parameters. For each one: the node as you configure it in n8n, what the parameter changes, and our field notes.

01

Model

model

What you see in n8n

Notes & use cases

This is the only parameter that changes what the vectors actually are. It picks the model which will generate the embeddings, and every other setting on the node only shapes how the request is sent.

Key parameters

  • Model: chosen from the option list in the node panel, and it has to stay identical between the run that indexes documents and the run that searches them.
Use cases
a knowledge base rebuilt from scratch is the right moment to change model, because every stored vector is replaced in the same pass and nothing is left behind in the old format.
02

Dimensions

options.dimensions

What you see in n8n

Notes & use cases

Vectors have a length, and this option sets it. Fewer dimensions means a smaller index and cheaper storage, more dimensions means finer distinctions between two similar paragraphs.

Key parameters

  • Dimensions: the number of dimensions the resulting output embeddings should have, picked from 256, 512, 1024, 1536 or 3072, and only supported in text-embedding-3 and later models.
Use cases
a product catalog where descriptions look alike benefits from a higher value, while an index of short support macros rarely needs one.
03

Base URL

options.baseURL

What you see in n8n

Notes & use cases

Requests go to OpenAI by default. Fill this field and they go somewhere else instead, which is what you need when the endpoint is not OpenAI itself.

Key parameters

  • Base URL: overrides the default base URL for the API, and the documentation points to it when you run a self-hosted OpenAI-like model.
Use cases
a company that keeps text inside its own network runs a compatible endpoint locally and points this field at it, so the documents never leave the building while the rest of the workflow stays unchanged.
04

Batch Size

options.batchSize

What you see in n8n

Notes & use cases

Documents are not sent one by one unless you ask for that. This number sets the maximum number of documents to send in each request, which decides how many round trips a large index build costs.

Key parameters

  • Batch Size: a number, raised to move a big import faster and lowered when the provider starts pushing back on large requests.
Use cases
indexing a long PDF split into hundreds of chunks is where the value matters; for a handful of chat questions it changes nothing you would notice.
05

Strip New Lines

options.stripNewLines

What you see in n8n

Notes & use cases

Line breaks carry no meaning for an embedding, but they do take up room in the text being sent. The toggle removes them before the request leaves n8n, and it is on by default.

Key parameters

  • Strip New Lines: whether to strip new lines from the input text, turned on unless you deliberately turn it off.
Use cases
text scraped from a web page arrives full of stray breaks, so leaving this on keeps the input clean; turn it off only if the layout of the text is part of what you want compared.
06

Timeout

options.timeout

What you see in n8n

Notes & use cases

A request that hangs will block the execution behind it. This value caps how long the node waits before giving up, expressed in seconds.

Key parameters

  • Timeout: the maximum amount of time a request is allowed to take in seconds, set to -1 for no timeout at all.
Use cases
an overnight index build can afford a generous value, while a chatbot answering a customer in a Slack thread should fail fast rather than leave the person waiting.
07

Encoding Format

options.encodingFormat

What you see in n8n

Notes & use cases

The vectors have to travel back over the network in some shape, and this selector decides which one. It affects the transport, not the meaning of the numbers.

Key parameters

  • Encoding Format: the format to return the embeddings in, either float or base64.
Use cases
leave it alone for a normal vector store setup, and reach for it only when the receiving side has a stated preference about how the payload arrives.
Need help

Need help automating Embeddings OpenAI with n8n?

A person reads every message.

FAQ

Questions people ask next

01Is the Embeddings OpenAI node included in n8n?
Yes. It ships inside the n8n AI package, @n8n/n8n-nodes-langchain, which comes with n8n. There is nothing to install, and n8n charges nothing extra for it, on n8n Cloud as well as on a self-hosted instance running the Community Edition under the Sustainable Use license. What is not free is the model side: OpenAI bills its own API calls under its own terms, exactly as it would if you called the API from anywhere else. n8n does not add a fee on top of that, and self-hosting the workflow does not change how the provider counts your usage.
02What do you need for it to work?
Two things. An OpenAI credential, created once from the Credentials menu in n8n and reusable across every workflow on the instance, and a root node to attach to. The node is a sub-node: it has no main input, it never runs by itself, and until its embeddings port is connected to something like a vector store, it simply does not execute. There is no Authentication selector on the node itself, so the credential is the whole of the setup. If you already created an OpenAI credential for a chat model, select that same one.
03What are the limits of the n8n Embeddings OpenAI node?
It does one thing and exposes 7 parameters to tune it, nothing more. The catalog lists a single version, version 1, so there is no older or newer variant with a different feature set. Splitting text into chunks is not its job, that belongs to a splitter node placed before the loader. There is no built-in retry, no caching of vectors between runs, and no way to mix two models in one index. Changing Model or Dimensions after documents are indexed means the old vectors no longer match the new ones, and a full rebuild is the only clean fix.
04What should you connect at a minimum to get started?
A root node that has an embeddings port, and this node underneath it. For a first build, a Simple Vector Store keeps everything in memory and gets you to a working search without any database to set up, which is the shape template 3986 uses. Once the idea holds, a Qdrant Vector Store is the step up, and templates 2846 and 2845 show it paired with an AI Agent and a question answering tool. Add the splitter and the data loader before the store, then keep the same model on both the indexing and the querying side.
05n8n or Make for this kind of AI workflow?
It depends on where the data has to live and how you want to pay. Make is hosted by Make, with no self-hosting option, and bills per operation, which is predictable but adds up fast when a single index build fires thousands of small calls. n8n runs on n8n Cloud or on your own server, so documents you vectorize can stay inside your own network, and the sub-node model here gives you direct control over the model and its options. Make is comfortable when the visual flow matters most and nothing has to stay in-house.
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