Resources · n8n integration

n8n Embeddings Google Vertex nodeConfigure Embeddings Google Vertex in n8n.

Text becomes numbers before a vector store can search it. The n8n Embeddings Google Vertex node turns text into embeddings, numerical vectors that let an AI compare meaning instead of exact words. It carries 2 parameters and plugs into one port. Built for anyone wiring a knowledge base into n8n.

Try the Embeddings Google Vertex node on n8nSee the 2 parameters

Verified Trustpilot reviews · AI, automation & growth agency

Why automate

What does the n8n Embeddings Google Vertex node actually do?

It is a sub-node, meaning a small node that never runs on its own and instead attaches to a bigger one. Its job is narrow: hand it text, and it calls Google Vertex to produce an embedding, a list of numbers that places that text somewhere in a map of meaning. A vector store, which is a database built to hold those numbers, then uses them to find passages that say the same thing in different words.

Say a support team keeps a few hundred answers in a document. Plain keyword search misses anything phrased differently. Route that text through this node and a vector store instead, and a question about a refund window finds the paragraph about returns even when the word refund never appears in it. Same embedding model on both sides, at indexing time and at question time, or the numbers stop lining up.

A second pattern: an internal assistant. The root node holds the conversation, this node quietly converts each incoming question into a vector, and the store returns the matching chunks before the answer is written. Teams often build that loop around the AI Agent node and hang the retrieved context off it.

Third, a classification job with no chat at all. Product notes arriving from a spreadsheet get embedded, compared against a handful of labelled examples already in the store, and written back with the closest label. Nothing here needs a chat model. It is arithmetic on meaning, and it runs fine in a scheduled workflow that reads from Google Sheets and posts the result to Slack.

When would you pick something else? If the account you already have is with another provider, the OpenAI side of the ecosystem covers the same job, and switching later means re-embedding everything you already stored. Choose once, deliberately. If your text never needs semantic search at all, skip embeddings and filter on fields.

The limits are worth knowing up front. This node exposes 2 parameters, Project ID and Model Name, and nothing else: no batching control, no dimension setting, no retry option in the panel. It has no main input, so you cannot test it by clicking run on the node alone. And Google bills the Vertex API calls it makes under your own contract; n8n adds nothing on top. Anyone who wants the wider picture of the platform can read the n8n review first.

Connect

How do you authenticate the node against Google Cloud?

  1. 01

    Create a Google service account credential

    Authentication runs through a Google service account, the credential type n8n documents for this node. In n8n, open the Credentials menu, create a new Google service account credential, and give it a name you will recognize later. A credential is stored once and reused by every workflow on the instance, so this step happens a single time even if you build several knowledge bases.

  2. 02

    Point the node at your project

    Back in the node panel, fill Project ID with the Google Cloud project that holds your Vertex access. The field is a resource locator, so n8n can list the projects the credential can see, or you can type the identifier by hand when the list is long. Pick the same project the service account belongs to, otherwise the call resolves against a project it has no rights on.

  3. 03

    Attach the node and run the parent

    Drag a connection from this node to the embeddings port of a root node, a vector store or an agent. Then run that parent node. Because a sub-node has no main input, the execution log shows its activity under the parent, not as a separate step. Seeing text go in and a long array of numbers come out is the sign that the credential works.

Connections

Where does this node plug in?

n8n splits AI work between root nodes, which sit in the main flow and receive items, and sub-nodes, which attach to a port on a root node and only wake up when that root node needs them.

Output (what it plugs into)

  • Embeddingsai_embedding
01EmbeddingsThis is the single port the node offers, and it is an output: the node does not accept connections, it supplies embeddings to whatever asks for them. Drop the node on the canvas and drag from its connector up to the embeddings slot of the parent.

This is the single port the node offers, and it is an output: the node does not accept connections, it supplies embeddings to whatever asks for them. Drop the node on the canvas and drag from its connector up to the embeddings slot of the parent.

A vector store in root mode is the usual parent, since it needs vectors both to write documents and to look them up. An AI Agent reaches the same vectors indirectly, through the store it queries rather than through a direct link here.

Key parameters

  • Required: a root node that expects embeddings cannot run without one attached, so the port is filled or the workflow stops.
Use cases
one node feeding one store keeps a knowledge base coherent, because everything inside it was measured with the same model.
Parameters

What do the 2 parameters control?

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

01

Project ID

projectId

What you see in n8n

Notes & use cases

Names the Google Cloud project the embedding calls are billed and authorized against. Without it the node has a credential but no idea which project to talk to, and the run stops before any text is converted.

Key parameters

  • Project ID: required, and built as a resource locator, so you either select your Google Cloud project from the list n8n pulls with the credential or enter the identifier directly.
Use cases
teams that keep a sandbox project next to a production one switch this field, and only this field, when a knowledge base graduates from testing to live.
02

Model Name

modelName

What you see in n8n

Notes & use cases

Chooses which Vertex model generates the vectors. Two documents embedded by two different models are not comparable, so this value is the one thing you should write down when you first index a corpus.

Key parameters

  • Model Name: a free text field, not required, which accepts a model identifier from the Google Vertex text embeddings API reference linked in the node description.
Use cases
changing the model on an index that already holds thousands of vectors means rebuilding the index, so test a new value on a copy of the store, never on the one an assistant is querying live.
Need help

Need help automating Embeddings Google Vertex with n8n?

A person reads every message.

FAQ

Questions people ask next

01Is the n8n Embeddings Google Vertex node included with n8n?
Yes. It ships inside the n8n AI package, so it is already there on n8n Cloud and on a self-hosted instance running the Community Edition under the Sustainable Use licence. There is no plugin to install and no extra fee on the n8n side, whichever way you host. What you do pay for sits elsewhere: Google bills the Vertex API calls the node makes, under your own agreement with them, and n8n adds nothing to that. A workflow built on n8n Cloud behaves identically once moved to a Docker install, which makes it safe to prototype in one place and run in the other.
02What do you need to make it work?
A Google service account credential, which is the authentication method documented for this node, plus a Google Cloud project identifier to fill Project ID. Create the credential once from the Credentials menu and every workflow on the instance can reuse it. Beyond that, you need a parent node: this is a sub-node with no main input, so it never appears as a step of its own in a workflow. Drag it onto the embeddings port of a vector store or a similar root node, and it starts working as soon as that node runs.
03What are the limits of this node?
It does one thing and exposes 2 parameters to do it, Project ID and Model Name. Version 1 gives you no batch size, no vector dimension setting, no timeout control from the panel. It also cannot be tested in isolation, since a sub-node without a main input has nothing to run on by itself, so every check happens through the parent node and its execution log. And it does not store anything: the vectors it produces live in whatever vector store you connected, which is the piece you configure for persistence, indexing and search.
04What should you connect at minimum to get started?
One root node and this one. Pick a vector store in root mode, attach this node to its embeddings port, and you have a working indexing pipeline: text goes into the store, vectors get written alongside it. From there, add retrieval. A second workflow queries the same store, and the answers feed an agent or a plain notification. Start with a small set of documents, a handful rather than an entire archive, and check that a question phrased in your own words returns the right passage before scaling the corpus up.
05n8n or Make for a Vertex embeddings workflow?
It depends on where you want the data to live. n8n self-hosts through Docker or npm, so the text you embed never leaves your own machine before the Vertex call, and the same workflow also runs on n8n Cloud unchanged. Make is hosted only, with no self-hosting option, and bills per operation, which changes the arithmetic when a workflow embeds documents in bulk. Visual logic differs too: n8n draws sub-nodes as separate pieces hanging off a root node, which suits AI work and makes a chain of embeddings, store and agent readable at a glance.
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