n8n AI Agents

Introduction

n8n is an open-source, low-code workflow automation platform. It uses a visual, node-based system to connect over 400 applications and APIs.

Its AI Agent functionality allows users to build autonomous workflows powered by Large Language Models (LLMs) that can make decisions, interact with apps, and execute complex, multi-step tasks without constant human input.

While standard n8n workflows follow a linear path (Trigger → Action → Action), the AI Agent functionality introduces autonomy.

What is an AI agent?

image-agents2

In n8n, an AI Agent is built visually. The AI Agent Node sits at the center, acting as the controller. You connect the “Brain” (LLM Model), “Tools,” and “Memory” directly to the agent node inputs. The workflow typically starts with a Trigger, such as a Chat Trigger or Webhook.

Instead of following a strict set of pre-defined steps, an Agent uses a Large Language Model (LLM) to:

Understand natural language input.
Reason about the best way to solve a problem.
Select Tools (such as search, database lookups, or API calls) to gather information.
Execute complex, multi-step tasks without constant human intervention.

image-agents

Accessing n8n

n8n is hosted on our Kubernetes/Rancher infrastructure.

Prerequisites

Account: A valid MetaCentrum account is required.
Platform: Access via Rancher https://rancher.cloud.e-infra.cz/.

Need an Account? If you are a member of an academic or research institution in the Czech Republic (within e-INFRA CZ) and do not have an account, please request a MetaCentrum account here.

For issues or assistance, please contact support at k8s@cerit-sc.cz.

Deployment guide: Start your own instance

Log in to Rancher

Navigate to the Rancher Dashboard https://rancher.cloud.e-infra.cz/dashboard/
Click Log in with Shibboleth
Select e-INFRA CZ AAI and choose e-INFRA CZ password among the organisations. Logi in using your MetaCentrum credentials.

Locate the n8n application

Once on the RANCHER dashboard

image-n8n1

Select KA —> Apps/Charts
Search for or select n8n

image-n8n2

Configure the instance

Click Install

image-n8n3

Name your instance: Enter a unique name for your deployment (e.g., username-n8n) and click Next. Fill in a name of the instance and remember it and click on Next

image-n8n4

Set the URL: You may enter a specific URL suffix or leave it empty for an automatically generated URL. Click Install.

image-n8n5

Finalize installation

Watch the installation logs. This process may take several minutes depending on available resources.

Important: When the installation completes, the logs will display your unique Access URL. Copy and save this URL immediately.

image-n8n6

First run configuration

Open your browser and navigate to the Access URL you saved in the previous step.
Create an Admin Account: You will be prompted to set up an email and password.

Note: This account is local to your n8n instance and is separate from your MetaCentrum login.

Once logged in, you will see the n8n canvas. You can now start building workflows.

image-n8n8

Managing resources — Terminating

To ensure fair usage and availability for all researchers, please manage your instance responsibly.

Deleting your instance

If you are no longer using n8n, please shut it down to free up resources.

Return to the Rancher Dashboard https://rancher.cloud.e-infra.cz/dashboard/ .
Locate your n8n instance in the list.
Click the three dots on the right side of the row.
Select Delete.

image-n8n8

AI model availability

We provide API access to high-performance models for your agents.

⚠️ Service Availability Notice Although Deepseek R1 0528 and Qwen 3 Coder are designated as guaranteed models, please be aware of the following infrastructure constraints:

Maintenance: The underlying hardware may require periodic maintenance. We currently lack backup infrastructure of comparable scale.
Priority Tasks: During urgent priority tasks requiring this hardware, model availability may be temporarily suspended.
Notifications: Advance notice of interruptions will be provided on this page and within the chat interface.

How to build an agent

Step 1: The trigger

To interact with your agent, you need a trigger. For testing and building chatbots, the Chat Trigger is the standard starting point.

For backend automation (like processing emails), you would use a Webhook or Email Trigger instead.

Step 2: Connecting the brain

Connect a Chat Model node (e.g., OpenAI) to the “Model” input of the Agent node. Our local models can be used .

Credential: You will need an API key from the provider
Model Selection: Choose a model capable of reasoning

Step 3: Configuring the agent

The AI Agent Node requires specific configuration to function correctly:

Setting	Recommendation
Agent Type	Select `Tools Agent`. This is the most versatile type that can use external tools.
Prompt Source	Usually `Connected Chat Trigger` or `Define below` if using a webhook.
System Message	CRITICAL. This is where you define the persona. Ex: “You are a helpful math assistant. Always use the calculator tool.”

Step 4: Empowering with tools

Without tools, an LLM is just a text generator. Tools give it hands. You can connect:

Pre-built Tools: Calculator, Wikipedia, Bing Search
Custom Tools: Any n8n workflow can be converted into a tool using the “Call Workflow” tool

Step 5: Managing memory

Window Buffer Memory: Keeps a “sliding window” of the last K messages. Best for: Keeping token costs low while maintaining immediate context.

Simple Memory: Stores the entire conversation history in execution RAM. Best for: Short, complex sessions where every detail matters.

Without a memory node connected, the agent treats every message as a brand new conversation.

Examples

(Step-by-step guides for common use cases will be added here soon.) TBD

n8n AI Agents

On this page