Resources · n8n integration

n8n Embeddings Azure OpenAI nodeConfigure Embeddings Azure OpenAI in n8n.

Every vector search starts with a model turning text into numbers. The n8n Embeddings Azure OpenAI node does that job through an Azure deployment, and exposes 5 options to control the model name, the batching, the text cleanup, the request timeout and the vector width. It is a sub-node, so it never runs alone.

Try the Embeddings Azure OpenAI node on n8nSee the 5 parameters

Verified Trustpilot reviews · AI, automation & growth agency

Why automate

What does the n8n Embeddings Azure OpenAI node actually do?

The node converts text into embeddings, meaning lists of numbers that place a sentence somewhere in a space where similar meanings sit close together. It calls a model deployment hosted on Azure OpenAI and hands the resulting vectors back to whatever root node asked for them, usually a vector store or a question answering chain. On its own it produces nothing: n8n only calls it when the node it is attached to needs text embedded.

The first scenario is the one most teams hit first: a knowledge base that has to stay current. A workflow pulls files from Google Drive, splits them, embeds each chunk through this node and writes the vectors into a store, which is exactly the shape of the published template 9174 built around a Simple Vector Store. When a document changes, the same run re-embeds it and the search index follows.

The second one is retrieval behind a chat surface. Template 6018 wires Slack to a Pinecone Vector Store with this node on the embeddings port, an AI Agent on top and an Auto-fixing Output Parser keeping the response structured. The question typed in Slack gets embedded by the same model that embedded the documents, which is what makes the comparison meaningful.

The third is a scheduled refresh of an internal catalog. Rows arrive from Google Sheets, the description column feeds the loader, and Batch Size decides how many of those rows travel in one request to Azure. Template 6020 follows that pattern for a Slack assistant.

When should you reach for something else? If your key is a plain OpenAI key rather than an Azure deployment, the OpenAI side of the ecosystem is the natural home. And if the task is generating an answer rather than measuring similarity, a chat model belongs on the root node, not this port: embeddings compare text, they do not write it.

Two limits are worth knowing before you build. Dimensions only applies to text-embedding-3 and later models, so an older deployment ignores the setting. And because this node has no main input, you cannot test it by clicking it alone; you run the root node and watch the embeddings port light up. Teams that want the whole pattern walked through end to end usually start with the n8n training.

Connect

What do you need before the node will run?

  1. 01

    Open an Azure OpenAI deployment

    The node talks to a model you have deployed on Azure OpenAI, not to a generic endpoint. Note the deployment name on the Azure side, because that exact string is what goes into Model (Deployment) Name later. Without a deployment there is nothing for the node to call, whatever the credential says.

  2. 02

    Create the credential once in n8n

    In n8n, a credential is a saved set of connection details that any workflow can reuse. Open the Credentials menu, create the Azure OpenAI entry and fill it from the official authentication page linked in the node documentation. Save it once and every embeddings node in every workflow can pick it from the dropdown.

  3. 03

    Attach the node to a root node

    Drag the node onto the canvas and connect it to the embeddings port of the node that needs vectors, a vector store or a chain. A port is the small socket under a root node that accepts one type of sub-node. Select your credential, then run the root node to see the first vectors come back.

Connections

Where does this node plug in?

n8n splits AI work between a root node, which receives the workflow items, and sub-nodes clipped onto its ports. This one is a sub-node with a single output port and no main input.

Output (what it plugs into)

  • Embeddingsai_embedding
01EmbeddingsThis is the socket the node hands its vectors to. A vector store or a chain asks for embeddings, and whichever embeddings sub-node sits on that port answers.

This is the socket the node hands its vectors to. A vector store or a chain asks for embeddings, and whichever embeddings sub-node sits on that port answers.

Key parameters

  • Required: the root node cannot store or search vectors without an embeddings sub-node on this port.
Use cases
a Pinecone Vector Store in insert mode filling an index, then the same store in query mode reading it back with the identical model on the port.
Parameters

What can you configure on the node?

The Embeddings Azure OpenAI node has 5 parameters. For each one: the node as you configure it in n8n, what the parameter changes, and our field notes.

01

Model (Deployment) Name

model

What you see in n8n

Notes & use cases

Picks which deployment generates the vectors. The value is the name of the model deployment to use, and it decides the shape and the behavior of everything downstream.

Key parameters

  • Model (Deployment) Name: a text value, typed or taken from the item with an expression such as {{ $json.model }} when a workflow serves several deployments.
Use cases
a document pipeline pinned to one deployment for the whole index, because vectors written by one model cannot be compared with vectors written by another.
02

Batch Size

options.batchSize

What you see in n8n

Notes & use cases

Sets the maximum number of documents sent in each request. Raise it and a long import finishes in fewer round trips; lower it and each call carries less.

Key parameters

  • Batch Size: a number, the ceiling per request rather than a fixed size, so a short run simply sends what it has.
Use cases
a nightly reindex of a help center, where hundreds of chunks go out in grouped requests instead of one call per paragraph.
03

Strip New Lines

options.stripNewLines

What you see in n8n

Notes & use cases

Removes new line characters from the input text before it reaches the model. n8n turns this on by default, which suits most prose.

Key parameters

  • Strip New Lines: a switch, on to strip the line breaks from the input text, off to leave them in place.
Use cases
indexing code snippets or tabular exports where the line breaks carry meaning, so the switch goes off to keep the original layout.
04

Timeout

options.timeout

What you see in n8n

Notes & use cases

Caps how long a single request is allowed to take, in seconds. Past that point the call stops instead of holding the execution open.

Key parameters

  • Timeout: a number of seconds, or -1 for no timeout at all.
Use cases
a chat assistant where a slow embedding call should fail fast and let the workflow answer, rather than leaving the person in Slack waiting.
05

Dimensions

options.dimensions

What you see in n8n

Notes & use cases

Fixes how wide the resulting vectors are. Narrower vectors take less room in the store; wider ones keep more nuance.

Key parameters

  • Dimensions: one of 256, 512, 1024, 1536 or 3072, supported only in text-embedding-3 and later models.
Use cases
matching the width your vector store index was created with, since an index built at one width rejects vectors of another.
Need help

Need help automating Embeddings Azure OpenAI with n8n?

A person reads every message.

FAQ

Questions people ask next

01Is the Embeddings Azure OpenAI node included in n8n?
Yes. The node ships inside the AI package of n8n, so there is nothing to install and nothing extra to pay on the n8n side, on n8n Cloud as well as on a self-hosted instance running the Community Edition under the Sustainable Use license. A workflow behaves the same in both places, which means you can build it locally and move it to Cloud later without touching the node. What does cost money is the model itself: the provider bills its own API calls under its own terms, and n8n adds nothing on top of that.
02What does the node need in order to work?
Two things. First, a credential for Azure OpenAI, created once in the Credentials menu of n8n and then reusable across every workflow you build. The node documentation links to the official authentication page for the exact values to fill in. Second, a root node to attach to, because this one is a sub-node: it has no main input and never runs on its own. Drop it on the canvas, connect it to the embeddings port of a vector store or a chain, then set the deployment name you created on Azure.
03What are the limits of the n8n Embeddings Azure OpenAI node?
The node does one thing: it turns text into vectors. It does not store them, search them or generate an answer, which are jobs for the vector store and the chat model sitting elsewhere in the workflow. The catalog lists version 1 of the node, with 5 options and no more. Dimensions only applies to text-embedding-3 and later models, so selecting a width on an older deployment changes nothing. And since there is no main input, you cannot execute the node by itself to test it.
04What do you plug in to get started?
At minimum, a root node that asks for embeddings and a credential on this node. The simplest first build is a vector store in insert mode with this node on its embeddings port and a data loader feeding it text, which is the pattern behind the Drive knowledge base template. Once that index exists, point a second flow at the same store in query mode and put the same model on the port. Add an AI Agent on top only when you want the answer written rather than the matches returned.
05n8n or Make for this kind of workflow?
It depends on where you want the data to live. n8n can be self-hosted with Docker or npm, or used on n8n Cloud, and the workflow is identical either way, so documents and vectors can stay on infrastructure you control. Make is hosted only, with no self-hosting option, and is billed per operation, which changes the arithmetic when a single reindex fires thousands of embedding calls. The visual logic differs too, so the honest test is to rebuild one small flow on both and see which one you read faster.
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