Resources · n8n integration

n8n NVIDIA Nemotron Chat Model nodeConfigure NVIDIA Nemotron Chat Model in n8n.

Nemotron runs your prompts, the agent does the rest. The n8n NVIDIA Nemotron Chat Model node is a sub-node, a small node you attach to a bigger one instead of running it alone. It exposes 9 parameters, from the model list to retries, and plugs into any root node that asks for a language model.

Try the NVIDIA Nemotron Chat Model node on n8nSee the 9 parameters

Verified Trustpilot reviews · AI, automation & growth agency

Why automate

What does the n8n NVIDIA Nemotron Chat Model node do in a workflow?

It gives another node its brain. On its own the node produces nothing: it has no main input and never runs by itself. You attach it to a root node, an AI Agent or a chain, through the port marked Model, and from then on every completion that root node needs is generated by a Nemotron model, either hosted on build.nvidia.com or served by a self-hosted NIM container on your own infrastructure.

Take a support assistant. An incoming message reaches an AI Agent, the agent decides what to answer, and the wording comes from the model you selected in Model. Keep Sampling Temperature low here: a support reply that changes tone on every run is a support reply nobody trusts.

Second case, product copy at volume. A workflow reads rows from Google Sheets, sends each row to a chain backed by this node, and writes the generated description back into the sheet. Maximum Number of Tokens is what stops one long-winded run from producing a paragraph where a line was expected, and Frequency Penalty keeps the model from repeating the same selling point on every row.

Third case, routing. A message lands in Slack, a classifier node decides which team owns it, and the answer has to be machine readable. Set Response Format to json_object, ask for JSON in the prompt of the chain or agent, and the next node can read the result with an expression such as {{ $json.category }}, the n8n syntax that pulls a field out of the item coming in.

When would you pick something else? The node only talks to Nemotron models through an OpenAI-spec compatible endpoint. If your account and your keys sit with another provider, the matching sub-node is the honest choice: OpenAI or Anthropic plug into exactly the same port, and swapping one chat model for another is a matter of dragging a connection. Staying on Nemotron makes sense when you want the same model available on build.nvidia.com and inside a NIM container you control.

The limits are worth knowing before you build. There is no main input, so a Nemotron node dropped on a blank canvas stays idle. n8n loads the model list from the endpoint configured in your credential, and falls back to a curated list of known Nemotron IDs when it cannot reach that endpoint. NVIDIA bills its own API calls under its own terms, n8n adds nothing on top. If you are still weighing the platform itself rather than the node, the n8n review goes through that question.

Connect

How do you authenticate the Nemotron node in n8n?

  1. 01

    Create the NVIDIA build account

    Cloud access starts with an account on build.nvidia.com. Open a Nemotron model in the catalogue, select Get API Key, and copy the key straight away. One credential covers both deployment modes, so this is also the account you keep if you later move part of the traffic to your own hardware. Nothing is installed in n8n at this stage.

  2. 02

    Add the credential in n8n

    In the Credentials menu, create an NVIDIA Nemotron credential. Two fields matter. Base URL is the OpenAI-spec compatible endpoint to call, and the default https://integrate.api.nvidia.com/v1 points at build.nvidia.com. API Key takes the key you just copied. Saved once, the credential is reusable across every workflow in the instance.

  3. 03

    Point it at a self-hosted NIM instead

    Running a NVIDIA Inference Microservice on your own infrastructure changes one thing: replace Base URL with your NIM endpoint, keeping the /v1 path, for example http://localhost:8000/v1. If that NIM enforces authentication, paste its token into API Key. If it does not, leave the field blank, which is allowed here and rare enough to be worth saying out loud.

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 it has a single outgoing connection.

Output (what it plugs into)

  • Modelai_languageModel
01ModelThis is the language model port, the socket a root node exposes when it needs text generated. Drag the Nemotron node onto it and the root node stops asking where its completions come from.

This is the language model port, the socket a root node exposes when it needs text generated. Drag the Nemotron node onto it and the root node stops asking where its completions come from.

Key parameters

  • Required: a root node without a chat model attached to this port cannot execute at all.
  • AI Agent and AI Agent Tool: the usual starting points, an agent that decides and acts.
  • Basic LLM Chain: one prompt in, one completion out, with no tool calling in between.
  • Text Classifier and Information Extractor: narrower root nodes that sort an item or pull structured fields out of free text.
Use cases
a triage workflow usually starts with Text Classifier on this port, then graduates to an AI Agent once the decisions stop fitting in a list of categories.
Parameters

What can you configure on the node?

The NVIDIA Nemotron Chat Model node has 9 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

Picks the Nemotron model that generates the completion. n8n loads the list dynamically from the endpoint set in your credential, and falls back to a curated list of well-known Nemotron IDs when that endpoint is unreachable.

Key parameters

  • Model: the dropdown itself, from the small nvidia/nvidia-nemotron-nano-9b-v2 up to nvidia/nemotron-3-super-120b-a12b, with reasoning and vision variants such as nvidia/nemotron-nano-12b-v2-vl in between.
Use cases
start a classification workflow on a nano model, and only move up the list if the answers are actually wrong.
02

Frequency Penalty

options.frequencyPenalty

What you see in n8n

Notes & use cases

Controls how often the model repeats itself. Positive values penalize tokens, the word fragments a model works with, based on how often they already appeared in the text, so the higher you go the less the model repeats the same line verbatim.

Key parameters

  • Frequency Penalty: a number you raise when generated text loops back on the same phrasing.
Use cases
generating several hundred product descriptions from one template, where the same closing sentence showing up everywhere is what gives the batch away.
03

Maximum Number of Tokens

options.maxTokens

What you see in n8n

Notes & use cases

Caps the completion length. The value is a token count, not a character count, and -1 hands the decision back to the model default.

Key parameters

  • Maximum Number of Tokens: leave it at -1 unless something downstream has a size limit to respect.
Use cases
a workflow writing into a field with a hard length limit, where a generous answer is worse than a short one. Watch the interaction with a low cap: the model does not summarize, it stops mid sentence, so give it room rather than trimming afterwards.
04

Response Format

options.responseFormat

What you see in n8n

Notes & use cases

Decides whether the node returns prose or parseable data. JSON mode should guarantee that what the model produces is valid JSON.

Key parameters

  • Text (text): a regular text response, the default behavior.
  • JSON (json_object): valid JSON, on one condition, the word json has to appear in the prompt of the chain or agent above.
Use cases
any branch where the next node reads a field rather than a sentence. Forget the word in the prompt and JSON mode is the option that silently disappoints.
05

Presence Penalty

options.presencePenalty

What you see in n8n

Notes & use cases

Pushes the model towards new subjects. Positive values penalize tokens that have already appeared at all, regardless of how often, which raises the chance the answer moves on rather than circling.

Key parameters

  • Presence Penalty: raise it for exploration, keep it down when the answer must stay on one topic.
Use cases
a brainstorming step that feeds a list of angles into the rest of the workflow. On a support reply it is the wrong knob, since wandering off topic is exactly the failure mode there.
06

Sampling Temperature

options.temperature

What you see in n8n

Notes & use cases

Sets how random the sampling process is. As the value approaches zero the model becomes deterministic and repetitive; a higher temperature creates more diverse sampling and increases the risk of hallucinations.

Key parameters

  • Sampling Temperature: the single most useful option on this node, and the first one to check when output quality moves around between runs.
Use cases
near zero for extraction and classification, higher for drafting. Changing this one and Top P together is generally not recommended.
07

Timeout

options.timeout

What you see in n8n

Notes & use cases

Sets the maximum time a single request is allowed to take, in milliseconds. Past that, the request is abandoned rather than left hanging.

Key parameters

  • Timeout: a number in milliseconds, so 2000 means two seconds, not two thousand.
Use cases
a self-hosted NIM that is still loading a large model answers slowly on the first call, and a tight timeout turns that warm-up into a failed execution. Long batches are where this option earns its place.
08

Max Retries

options.maxRetries

What you see in n8n

Notes & use cases

Says how many times n8n retries a request before giving up on it.

Key parameters

  • Max Retries: the retry count for one request, worth raising on a shared endpoint and worth lowering when a failure should surface fast.
Use cases
a nightly run over a long list, where one refused request should not stop the whole execution. Pair it with Timeout: retries on top of a generous timeout can keep a workflow busy far longer than expected.
09

Top P

options.topP

What you see in n8n

Notes & use cases

Controls diversity through nucleus sampling. At 0.5, half of all likelihood-weighted options are considered, so a lower value makes the model ignore its less probable choices.

Key parameters

  • Top P: the probability mass the completion is allowed to draw from.
Use cases
tightening answers on a workflow that already runs at a fixed temperature. Altering this option or Sampling Temperature is generally recommended, but not both at once, otherwise no run tells you which knob did what.
Need help

Need help automating NVIDIA Nemotron Chat Model with n8n?

A person reads every message.

FAQ

Questions people ask before building

01Is the n8n NVIDIA Nemotron Chat Model node included in n8n?
Yes. It ships inside the n8n AI package, @n8n/n8n-nodes-langchain, which comes with n8n itself. There is nothing to install and nothing extra to pay on the n8n side, whether you run n8n Cloud, the hosted offer, or a self-hosted instance under the Community Edition and its Sustainable Use licence. A workflow behaves the same way in both. What you do pay is NVIDIA: the model provider bills its own API calls under its own terms, and n8n adds nothing on top of that. Pointing the node at a NIM container on your own infrastructure changes where the compute happens, not what n8n charges.
02What do you need to make it work?
One credential, created once in the Credentials menu and reused in every workflow. It holds a Base URL, the OpenAI-spec compatible endpoint to call, and an API Key. For build.nvidia.com, keep the default endpoint and generate a key from the model catalogue with Get API Key. For a self-hosted NIM, replace the Base URL with your own endpoint including the /v1 path, and leave the API Key blank if that NIM does not enforce authentication. There is no Authentication selector on the node itself. Beyond the credential, the node needs a root node to attach to, since a sub-node never executes on its own.
03What are the limits of this node?
Three worth knowing. It has no main input, so it cannot be triggered, cannot sit between two regular nodes, and does nothing until a root node asks it for a completion. It only reaches Nemotron models through an OpenAI-spec compatible endpoint, which means build.nvidia.com or a NIM you run yourself, and nothing else. And the Model dropdown depends on that endpoint being reachable: when it is not, n8n shows a curated list of well-known Nemotron IDs instead of what your deployment actually serves. The node is at version 1, so an older workflow shows the same options.
04What is the minimum you have to plug in, and where should you start?
One connection, on the Model port. Attach the node there and the root node has everything it needs. To start, Basic LLM Chain is the gentler option: one prompt goes in, one completion comes out, with no tool calling to debug alongside the model. Once the workflow needs decisions rather than a single answer, move to AI Agent, which accepts the same port. Text Classifier and Information Extractor are good intermediate steps when the job is sorting items or pulling fields out of free text. Guardrails and Summarization Chain accept this port too.
05n8n or Make for running Nemotron models?
It depends on where you want the workflow to live. Make is hosted only, with no self-hosting option, and bills per operation. n8n runs on its cloud or on your own servers, which is the deciding factor here: if you are using Nemotron through a NIM container precisely to keep inference on your own infrastructure, sending every prompt through a hosted platform works against that choice. Make has a visual logic many teams find quicker to read at a glance. Cost models differ too, per operation on one side, per instance on the other, so run your own volumes through both before deciding.
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