n8n AI Automation Tutorial: Complete Workflow Setup Guide (2026)

n8n (pronounced “nodemation”) is an open-source workflow automation tool. It connects apps, moves data between them, and automates repetitive tasks without much code.

Instead of manually copying leads from a form into your CRM, writing follow-up emails, and updating a spreadsheet, n8n does all of that automatically — every time, the moment a form is submitted.

What changed in 2026 is AI. You can drop an AI agent powered by GPT-4o, Claude, or Gemini directly inside a workflow. The agent reads context, makes decisions, calls APIs, and writes responses. This isn't a bolt-on summarizer. AI is part of the automation itself.

n8n 2.0 launched in January 2026 with native LangChain integration, 70 AI-dedicated nodes, sandboxed code execution, persistent agent memory, and native MCP support.

Two things shifted n8n's trajectory. n8n 2.0 moved it from “developer-friendly Zapier clone” to a real AI orchestration engine. The other was cost.

Zapier charges per task, and every action in a workflow counts separately. A 10-step workflow running 1,000 times a month burns through 10,000 tasks. A self-hosted n8n instance running the same volume on an ₹800/month VPS still costs ₹800/month — no execution limits, no per-task charges, same software.

Zapier Professional runs roughly ₹4,720/month with GST for 2,000 tasks. A self-hosted n8n on a basic DigitalOcean droplet handles most startup automation at ₹800–1,500/month total. Past 10,000 runs a month, the economics aren't close.

If you're a non-technical team running simple tasks across standard SaaS tools, Zapier wins on ease. For developer-led teams doing AI-heavy work at volume, or with data residency requirements, n8n is the one to use.

Option A: n8n Cloud. Go to n8n.io, sign up, and you get a managed instance. There's a 14-day free trial. Starter is about ₹1,700/month for 2,500 executions. Good for testing. Expensive for production.

Option B: Self-hosting with Docker. Recommended. You need a Linux server. For Indian founders: DigitalOcean basic droplet (₹800–1,500/month), Hetzner Cloud (cheaper but Europe), or AWS Mumbai ap-south-1 (pricier, keeps data in India). Minimum: 2 vCPU and 4 GB RAM.

services:
  postgres:
    image: postgres:16
    restart: unless-stopped
    environment:
      POSTGRES_USER: n8n
      POSTGRES_PASSWORD: change_this_password
      POSTGRES_DB: n8n
  n8n:
    image: docker.n8n.io/n8nio/n8n
    restart: unless-stopped
    ports:
      - "5678:5678"
    environment:
      - DB_TYPE=postgresdb
      - DB_POSTGRESDB_HOST=postgres
      - N8N_ENCRYPTION_KEY=a_random_32_char_key
      - EXECUTIONS_DATA_MAX_AGE=168
      - N8N_EDITOR_BASE_URL=https://n8n.yourdomain.com

Two things worth flagging: use PostgreSQL from day one — migrating from the default SQLite later is painful. And set EXECUTIONS_DATA_MAX_AGE=168 (7 days) so execution logs don't balloon.

Add Caddy as a reverse proxy for automatic HTTPS certificates. Don't expose port 5678 directly to the internet.

The fastest way to understand n8n is to build something real. This workflow watches Gmail for new emails, summarizes them with AI, and posts the summary to Slack. It covers the three building blocks behind most n8n workflows: a trigger, a transformation, and an action.

1. Gmail Trigger. Connect via OAuth2, set event to “New Email Received,” poll every 5 minutes.
2. OpenAI node. Resource: Chat, Model: gpt-4o. Prompt: “Summarize this email in exactly 3 bullet points. Focus on what action is required, who it's from, and any deadline mentioned. Subject: {{ $json.subject }}”
3. Slack node. Connect your workspace, operation: Send message, channel: #email-summaries.
4. Test it. Click “Test Workflow.” n8n pulls a real email, runs OpenAI, posts to Slack. You see data flowing through each node in the execution panel.
5. Activate. It runs automatically in the background.

The {{ $json.field }} syntax is how n8n passes data between nodes. The Gmail trigger outputs the email as JSON, and this expression pulls specific fields into your prompt. Once that clicks, the rest of the system makes sense.

Trigger nodes start a workflow. They listen for something — a new email, a webhook call, a schedule, an incoming chat message.

Action nodes connect to external services — Google Sheets, Slack, Razorpay, HubSpot, Notion. n8n ships with 500+ built-in integrations. The HTTP Request node covers anything not natively supported.

The AI Agent node is the most powerful primitive in 2026. It's not a simple GPT wrapper; it's an orchestration layer connecting an LLM to a set of tools. The agent reads input, decides which tool to call, calls it, reads the result, reasons about what to do next, and loops until it has enough to respond. This is the ReAct (reason and act) pattern as a visual node.

We're building a customer support agent that looks up orders, answers FAQs, and escalates to a human when it can't resolve something. This covers the full AI Agent pattern.

Step 1: Add a Chat Trigger. It creates a built-in interface for testing and generates a sessionId for multi-turn conversations.

Step 2: Add an AI Agent node with a system prompt defining the agent's rules — which tools it has, when to escalate, how concise to be, and critically: never make up data it doesn't have.

Step 3: Attach an OpenAI Chat Model sub-node. Use gpt-4o for complex queries and gpt-4o-mini for simple FAQ lookups (more on routing below).

Step 4: Memory. Use Postgres Chat Memory for production, connected to your PostgreSQL instance with the session key set to {{ $('Chat Trigger').item.json.sessionId }}. Simple Memory (RAM) is lost on every restart.

Step 5: Tools. Add HTTP Request nodes as tools. The key syntax is $fromAI('identifier', 'the order ID or customer email'). This tells the agent what data to extract and pass at runtime — the agent fills in these values based on its own reasoning.

Step 6: Add an IF node after the agent to catch escalation intent and route true to a Slack notification for your support team. Every tool call and result is visible in the execution log — one of n8n's biggest practical advantages over a custom-coded agent pipeline.

Most global n8n tutorials skip this entirely. If you're building in India — fintech, healthtech, edtech, SaaS — data residency isn't optional. The Digital Personal Data Protection (DPDP) Act 2023, with rules notified in November 2025, applies to any organization processing digital personal data of Indian citizens. Full compliance obligations land by May 2027. The Consent Manager Framework becomes operational by November 2026.

Zapier and Make store your workflow data on servers in the US or EU. Every customer record, API response, and email payload running through Zapier leaves India. Self-hosted n8n on Indian infrastructure keeps data in India — AWS Mumbai (ap-south-1), a DigitalOcean Bangalore droplet, or a managed Indian provider.

For fintech and healthtech handling Aadhaar data, PAN, bank statements, or health records, data residency makes self-hosted n8n the only option that holds up under DPDP regardless of volume.

MCP (Model Context Protocol) is an open standard letting AI agents connect to external services through a standardized interface. n8n shipped native MCP support in May 2026.

Before this, connecting an agent to Notion, Linear, or Monday.com meant manually configuring an MCP Client node with credentials, endpoint URLs, and auth flows. Now you pick the service from the nodes panel, authenticate once, and it's available as a tool.

Current native MCP coverage includes Apify, Linear, Monday.com, Notion, and PostHog, with more being added continuously. If an agent needs to create Notion pages, update Linear issues, or read PostHog analytics, that's a two-minute addition now — no HTTP node configuration needed.

By default, n8n runs all workflow executions in a single process. Under concurrent load — multiple workflows triggering at once, long-running AI agent calls — this becomes a bottleneck.

The fix is queue mode with Redis. It separates the main process (UI and API) from worker processes, which scale horizontally. Add Redis to your docker-compose.yml, set EXECUTIONS_MODE=queue on both the main instance and worker containers, and add more worker replicas as needed — each pulls jobs from the Redis queue independently.

n8n is workflow orchestration. It handles triggers, data flow, API calls, and LLM invocation. Here's what it doesn't handle.

Teams that struggle with n8n in production usually treated it as a magic box. The ones that succeed treat it like any other production infrastructure: documentation, error handling, monitoring, clear ownership.