Resources · n8n integration

n8n Google Vertex Chat Model nodeConfigure Google Vertex Chat Model in n8n.

The n8n Google Vertex Chat Model node is the part that actually talks to Google's models. It is a sub-node, meaning it never runs on its own: it clips onto the Model port of a root node such as an AI Agent. Eight parameters, from Project ID to Thinking Budget, decide what comes back.

Try the Google Vertex Chat Model node on n8nSee the 8 parameters

Verified Trustpilot reviews · AI, automation & growth agency

Why automate

What does the n8n Google Vertex Chat Model node do?

It gives a root node its brain. In n8n, an AI Agent or a chain handles the workflow logic, but it produces nothing until a chat model is attached to it, a chat model being the component that turns a prompt into text. This node is that component, backed by Google Vertex AI. You pick a Google Cloud project, name a model, and adjust how the answer is generated.

First setup, support triage. An AI Agent reads incoming messages pulled from Gmail, decides which team owns each one, and drafts a reply. The agent holds the instructions, this node holds the model: set Model Name to a fast Gemini model and the whole loop stays cheap enough to run on every message.

Second setup, structured extraction. A Basic LLM Chain reads a free-text form answer and returns clean fields, which a later node writes into Google Sheets. Here Sampling Temperature matters more than the model: push it toward zero and the same input gives the same columns run after run.

Third setup, a daily digest. A chain summarizes yesterday's tickets and posts the result to Slack. Maximum Number of Tokens is the lever that keeps the digest readable instead of a wall of text, since it caps how long the completion can get.

When to pick something else. If the account already runs on an OpenAI key, the OpenAI chat model plugs into the exact same port with one credential less to manage. Vertex earns its place when the data already sits in Google Cloud and the project, quotas and billing are meant to stay there. Both are sub-nodes, so swapping one for the other is a matter of dragging a different box onto the Model port.

Known limits. This node exposes the parameters listed by the catalogue at node version 1 and nothing more: project, model, four sampling controls, safety filters and a thinking budget. It has no main input, so it never appears in the middle of a workflow line. And the model provider bills its own API calls under its own terms, whatever n8n does around it. That last point is also worth weighing in the n8n review before standardizing a stack on it.

Connect

What do you connect before the node answers?

  1. 01

    Open the Google Cloud project

    The node asks for a Project ID, so the Vertex AI side has to exist first. Whoever owns the Google Cloud account picks the project that will carry the model calls and their billing. That identifier is what n8n sends with every request, and it is the value you will select or type in the first parameter of the node.

  2. 02

    Create the credential in n8n

    Authentication goes through a Google service account credential, as pointed out by the node documentation. In n8n, open the Credentials menu, create it once, fill in the service account details, and save. The credential is stored at account level, so every workflow that needs a Vertex model reuses it instead of asking for keys again.

  3. 03

    Attach the node to a root node

    Drag the node onto the canvas and connect it to the Model port of an AI Agent, a chain or another root node. Once the credential is selected, the Project ID list loads from the Google Cloud account, which is the fastest way to check the connection works. Then set Model Name and run the parent node once.

Connections

Where does this node plug in?

n8n splits AI work in two: a root node receives the workflow items and drives the logic, and sub-nodes clip onto its ports to supply a model, a memory or a tool. This node only ever fills the model slot.

Output (what it plugs into)

  • Modelai_languageModel
01ModelThis is the single port the node exposes, and it is an output: the connection runs from the chat model up into the node that will use it.

This is the single port the node exposes, and it is an output: the connection runs from the chat model up into the node that will use it.

Key parameters

  • Required: a root node with a Model port produces no text until something fills it, so the wire is not optional.
  • AI Agent: takes it when the workflow needs decisions and tool calls, and so does the AI Agent Tool variant.
  • Basic LLM Chain: takes it for a single pass over a prompt, like Question and Answer Chain over a document.
  • Auto-fixing Output Parser: accepts one too, using the model a second time to repair malformed output.
Use cases
the same Vertex model can feed an agent and, further down the canvas, the parser that cleans up what that agent returned.
Parameters

What does each parameter change?

The Google Vertex Chat Model node has 8 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 calls are charged and routed to. It is the only required parameter of the node, and nothing runs while it is empty.

Key parameters

  • Project ID: select or enter your Google Cloud project ID. n8n loads the list of projects from the connected Google Cloud account, and manual entry stays available when the list is long or the account holds many projects.
Use cases
teams that keep a sandbox project and a production project point staging workflows at one and live workflows at the other, without touching anything else in the node.
02

Model Name

modelName

What you see in n8n

Notes & use cases

Says which model generates the completion. Change this one line and the same agent suddenly answers faster, or thinks harder.

Key parameters

  • Model Name: the model which will generate the completion, entered as a string, for example gemini-1.5-flash-001 or gemini-1.5-pro-001. Google's model list is the reference for what the project can call.
Use cases
a high-volume classification step runs on a flash model, while the drafting step at the end of the same workflow gets a pro model. Two nodes, two names, one credential.
03

Maximum Number of Tokens

options.maxOutputTokens

What you see in n8n

Notes & use cases

Caps the completion length. A token is roughly a word fragment, so this is the ceiling on how much text comes back, not on what goes in.

Key parameters

  • Maximum Number of Tokens: the maximum number of tokens to generate in the completion. Leave it alone for open-ended drafting, lower it when the answer feeds a field with a size limit.
Use cases
a summary posted as a chat message is worth capping, because a model that starts listing every detail turns a digest into something nobody reads to the end.
04

Sampling Temperature

options.temperature

What you see in n8n

Notes & use cases

Decides how adventurous the model is when choosing its next word. Lower means less random completions, and as the temperature approaches zero the model becomes deterministic and repetitive.

Key parameters

  • Sampling Temperature: controls randomness. Higher values create more diverse sampling but increase the risk of hallucinations, where the model states something confidently wrong.
Use cases
anything parsed by a later node wants a low value, so two identical inputs give two identical outputs. Copy variations for a newsletter want the opposite.
05

Top K

options.topK

What you see in n8n

Notes & use cases

Trims the pool of candidate tokens before the model picks one, which removes the long tail of low probability responses.

Key parameters

  • Top K: the number of token choices the model uses to generate the next token. It defaults to -1, which disables it entirely, so the parameter does nothing until a value is set.
Use cases
a model that occasionally drops an odd word into otherwise clean output gets tidier once the tail is cut, without touching how creative the rest of the answer is.
06

Top P

options.topP

What you see in n8n

Notes & use cases

Another way to narrow the same pool, this time by probability mass rather than by count. It is called nucleus sampling.

Key parameters

  • Top P: controls diversity via nucleus sampling, where 0.5 means half of all likelihood-weighted options are considered. The general recommendation is to alter this or the temperature, but not both.
Use cases
when an agent's answers drift and the temperature has already been lowered without result, this is the second dial to try. Turning both at once makes the behavior impossible to reason about.
07

Safety Settings

options.safetySettings

What you see in n8n

Notes & use cases

Adjusts the filters Google applies to the generated content. The parameter is a collection: click Add Option and each entry pairs a category with a threshold.

Key parameters

  • Safety Category: the category of harmful content to block, among HARM_CATEGORY_HARASSMENT, HARM_CATEGORY_HATE_SPEECH, HARM_CATEGORY_SEXUALLY_EXPLICIT and HARM_CATEGORY_DANGEROUS_CONTENT.
  • Safety Threshold: the threshold of harmful content to block, from BLOCK_LOW_AND_ABOVE down to BLOCK_ONLY_HIGH, with BLOCK_NONE allowing all content and HARM_BLOCK_THRESHOLD_UNSPECIFIED leaving it unspecified.
Use cases
a workflow that summarizes user-reported abuse gets blocked by default filters on its own subject matter, and needs a looser threshold on the matching category to work at all.
08

Thinking Budget

options.thinkingBudget

What you see in n8n

Notes & use cases

Governs how much internal reasoning a thinking model is allowed before it answers. Those reasoning tokens are counted like any other.

Key parameters

  • Thinking Budget: set it to 0 to disable automatic thinking, to -1 for dynamic thinking, and leave it empty for auto mode. It only bites on models that support thinking.
Use cases
a routing step that only has to choose between a few labels rarely benefits from extra reasoning, so disabling it keeps latency down on the busiest branch of the workflow.
Need help

Need help automating Google Vertex Chat Model with n8n?

A person reads every message.

FAQ

Questions that come up next

01Is the Google Vertex Chat Model node included in n8n?
Yes. It ships inside n8n's AI package, 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 licence. A workflow behaves the same in both places. The one cost that does exist sits elsewhere: the model provider bills its own API calls under its own terms, and n8n adds nothing on top of that. So the node itself is free to use, while the answers it generates are charged by Google against the project you named in Project ID.
02What do you need to make it work?
A Google Cloud project and a credential. The node documentation points to a Google service account for authentication, which you create once in the n8n Credentials menu and then reuse across every workflow. On top of that, you need a root node to attach it to, because this is a sub-node: it has no main input and never executes by itself. Attach it to the Model port of an AI Agent or a chain, select the credential, then fill in Project ID and Model Name. Those two parameters are the only ones you have to touch before a first run.
03What are the limits of the n8n Google Vertex Chat Model node?
It does one job and exposes eight parameters to tune it: the project, the model name, then the maximum number of tokens, the sampling temperature, Top K, Top P, the safety settings and the thinking budget. Anything outside that list is not configurable from the node. It also cannot sit in the middle of a workflow line, because it has no main input, and it produces nothing when executed alone. The page describes node version 1 as read in the catalogue, so an older workflow may show a version with fewer options than the ones listed here.
04What should you connect at minimum, and which root node to start with?
At minimum, one wire and one credential. The wire goes from this node to the Model port of a root node, and the credential gives it access to Vertex AI. Nothing else is mandatory: memory, tools and output parsers are separate sub-nodes you add later, each on its own port. To start, a Basic LLM Chain is the gentler option, since it sends one prompt and returns one answer, which makes it obvious whether the model is reachable. Move to an AI Agent once the workflow needs the model to pick between actions rather than just write text.
05n8n or Make for this use case?
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 it bills per operation, which suits teams that want nothing to run. n8n can be self-hosted or used on n8n Cloud, with the same workflow either way, which matters when the Google Cloud project and its data are meant to stay under your own control. The visual logic also differs: n8n treats the AI model as a sub-node you attach to a root node, so the model becomes a swappable part rather than a setting buried in a step.
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