"AI Automation 2026 — n8n + AI Nodes, Three Workflow Recipes"

Compared with Zapier and Make, self-hosting options, OpenAI/Claude wiring, and three copy-pastable flows

핵심 요약

  • Audience: Anyone losing time on repetitive tasks — solo operators, small teams, plus developers automating intake, content, or support.
  • What you'll get: 1) How n8n differs from Zapier/Make, 2) cloud vs. self-hosted pricing, 3) three immediately reproducible recipes (email triage, content syndication, customer-support draft), 4) wiring OpenAI/Claude, 5) limits and cost controls.
  • Core idea: n8n is a workflow tool with native AI agent nodes. Steeper learning curve than Zapier, but unbeatable on cost, scale, and self-hosting.

1. What n8n is

n8n = a workflow automation platform. Triggers (email, webhook, schedule) flow through nodes (apps, logic, LLMs) to run on autopilot via a visual canvas. The native "AI Agent" node lets you build LangChain-style agents inline.

How it differs from Zapier and Make

n8n Zapier Make
Integrations 400+ 8,000+ 1,500+
Pricing model Per workflow run Per task (per node) Per operation
Self-hosting ✅ Free Community Edition ❌ ❌
Cloud entry €24/mo (2,500 runs) $19.99/mo (750 tasks) $9/mo (10,000 ops)
AI nodes 70+ dedicated + LangChain integration Copilot + AI Agents (beta) AI Agents (Oct 2025)
Learning curve Moderate Easy Moderate

Sources: n8n pricing, Zapier vs n8n.

One-liner

  • Zapier = easiest, but expensive at scale (you pay per step).
  • Make = visual, balanced.
  • n8n = most powerful and cheapest (especially self-hosted), with a slightly steeper curve.

2. Pricing — which plan to pick

Cloud

Plan Price Runs Best for
Starter €24/mo 2,500/mo Solo, light usage
Pro €60/mo 10,000/mo Steady production
Business €800/mo 40,000/mo + SSO SMBs
14-day trial – – No credit card

Self-Hosted (Community Edition)

  • Free, unlimited runs.
  • You pay infrastructure (€5–200/mo, but a €5 VPS often covers a single user).
  • Docker install ~5 minutes (official guide).

Which to pick

Situation Recommendation
Just experimenting Cloud Starter, 14-day free trial
Under 2,500 runs/month Cloud Starter (€24)
~10K runs/month + reliability Cloud Pro (€60)
Data-sensitive / unlimited runs Self-hosted Community
Team / SSO Cloud Business or self-hosted Enterprise

Common path: start cloud for a week, then move to self-hosting once your workflows stabilize.

3. AI nodes — where n8n meets LLMs

3.1 Supported LLM providers

  • OpenAI: GPT-5.5, GPT-5.4, GPT-4o
  • Anthropic: Claude Opus 4.7, Sonnet 4.6
  • Google: Gemini 3 Pro, Flash
  • Local LLMs: Ollama (Llama 3, Mistral, Qwen)
  • Others: HuggingFace, Cohere

3.2 Four core nodes

Node Role
AI Agent LLM + tools (other nodes) + memory. ReAct pattern.
OpenAI / Anthropic / Gemini Chat Plain LLM call (input → output)
Memory Window / Buffer / Vector — retain conversation context
Tool Functions an agent can call (HTTP, Sheets, Slack, etc.)

3.3 Cost model — n8n itself + LLM API

  • n8n executions: 1 run = 1 execution. A 20-step workflow still counts as 1 execution.
  • LLM API: billed separately (your OpenAI / Anthropic / Google key).
  • Example: email-triage workflow at 100/day → 3,000 runs/month = Pro plan + Claude Haiku tokens (~$2–5/month).

4. Three copy-pastable recipes

4.1 Recipe 1 — Email triage and summary

Problem: Inbox is on fire. You need important / to-do / newsletter / spam triage daily.

Workflow

Gmail trigger (new message)
  → Claude Haiku node (classification prompt)
    "Classify this email as one of [Important / To-do / Newsletter / Spam]
     and add a one-line summary if it's Important or To-do."
  → Switch (route by label)
    - Important → Slack DM
    - To-do → Notion database row
    - Newsletter → Just label it
    - Spam → Trash

Estimated cost - 100 messages/day × 30 days = 3,000 runs → Pro €60/mo - Haiku tokens (~200 in × 100/day) ≈ $0.05/day → ~$1.50/month - Total ≈ €60 + $1.50 ≈ €61.50/month

Alternative: self-host + Haiku → ~$1.50/month total.

4.2 Recipe 2 — Write once, distribute everywhere

Problem: You write a blog post and don't want to re-edit it for LinkedIn, Twitter, and Discord every time.

Workflow

Webhook trigger (receives blog post URL)
  → HTTP Request (fetch the article)
  → AI Agent (Claude Sonnet 4.6, 4-block system prompt)
    Output JSON: { linkedin_post, twitter_thread, discord_summary }
  → Three branches:
    - LinkedIn API → publish (or save draft)
    - Twitter API → post thread
    - Discord webhook → channel summary

Trick: one LLM call returns JSON for all three channels — token cost ⅓ vs three separate calls.

4.3 Recipe 3 — Customer-support first-pass drafts

Problem: Your small business gets daily inquiries you can't answer within 24 hours.

Workflow

Gmail trigger (specific label or main inbox)
  → AI Agent (Claude Sonnet 4.6 + Notion DB tool)
    Tool 1: Search Notion FAQ
    Tool 2: Lookup order DB
    Prompt: "Read the customer email, find the answer in the FAQ;
     if it's order-related, query the order DB.
     If you're unsure, return 'NEED_HUMAN'."
  → If NEED_HUMAN
    Yes → Slack #cs channel for human review
    No → Save Gmail *draft* (recommend manual click before sending)

Safety rails - Start with drafts only, not auto-send. - Monitor accuracy for a week before allowing auto-send for safe categories. - Keep the "NEED_HUMAN" threshold conservative.

5. Limits + cost control

Limits

  • Hallucination: Without RAG or Tool Use grounding, LLMs hallucinate.
  • Legal liability: Auto-sending email replies, contracts, or payments needs human review.
  • Failure cascades: One API outage stops the whole workflow. Build retries and dead-letter queues.
  • Secrets exposure: Manage API keys via n8n Credentials, not in node JSON.

Five cost controls

  1. Tier the model: Haiku/GPT-4o-mini for classification, Sonnet/Opus for reasoning. ~70% cost cut.
  2. Pre-filter with Switch: Cheap rules handle 80% of branches before you ever call an LLM.
  3. Cap memory: Use Window Memory (last N) instead of full Buffer Memory.
  4. Reduce schedule frequency: 5 min → 15 min cuts run count to ⅓.
  5. Self-host + free LLM: Self-hosted n8n + Ollama Llama 3 ≈ free aside from your VPS bill.

6. First three days

Day Task
1 Sign up for n8n Cloud Starter 14-day trial. Build one workflow without AI: "RSS feed → Slack."
2 Get an OpenAI or Anthropic API key. Add an AI Agent node. Build Recipe 1 (email triage).
3 Evaluate self-hosting (€24/mo cloud vs. €5 VPS + free n8n). Pick 1–2 workflows to operationalize.

7. What it isn't

  • Not a BI tool → Metabase / Looker.
  • Not an ML pipeline orchestrator → Airflow / Prefect.
  • Not for accounting/payment auto-execution → talk to Stripe / QuickBooks directly.
  • Not real-time chatbot infra → host a frontend on Vercel and call LLMs there.

n8n's correct slot: glue between SaaS, LLMs, and HTTP endpoints, with a visual canvas.

Developer notes

  1. Self-host with Docker: docker run -it --rm -p 5678:5678 n8nio/n8n (use a PostgreSQL backend in production).
  2. Backups: SQLite is fine for tiny workflows. For real use, PostgreSQL + automated backups.
  3. Version control: Export/import workflows as JSON, store in Git, review in PRs.
  4. Error workflows: Build a dedicated Error Workflow that other workflows reference, alerting Slack/Discord on failure.
  5. WebSockets + webhooks: n8n exposes both — useful for real-time chatbot frontends.
  6. MCP support: n8n 2.x is shipping Model Context Protocol support — call standardized external LLM tools.
  7. AI Agent node vs raw LangChain: simple flows → AI Agent node; complex graphs → LangGraph plus an n8n HTTP trigger.

References

This is part 7 of 11 in the AI Basics series. Next: n8n in practice — Discord/Notion/Gmail integrations (extension).

댓글

댓글 쓰기

이 블로그의 인기 게시물

Agent Memory Engine (2/10) — Building an AI Agent Memory System with SQLite Alone

← 1/10 SQLite Memory Engine f… 📚 Series Index 3/10 memcore Completed → A memory engine that runs without daemons, without dependencies, anywhere 핵심 요약 This post covers how to design an AI agent memory engine (memcore) on a single-file SQLite backend. - Structure: 25 modules, ~6,464 lines, 22 DB tables - Principles: no daemon, single-file portability, no required external dependencies, host-agnostic - Core techniques: 4-layer hybrid search (topic · vector · FTS5 · LIKE), U-tag dialectic-based confidence evolution, ingestion-layer bias rejection rules What This Post Covers One approach to building AI agent memory without a server or cloud vector DB. Specifically: (1) the limitations of text-file-based memory, (2) vector DB alternatives compared and selection criteria, (3) how to layer search tiers, (4) how to evolve confidence scores over time, and (5) how to block biased memory accumulation at the ingestion stage. Design Principles: Why Single-File SQLite Text-file-based ...
Read more »

"ML Foundations (9/9) — PyTorch vs TensorFlow, and the Road to Local LLMs"

이미지
← 8/9 Deep Learning Architec… 📚 Series Index (series end) The final part. So far we've focused on models. Now we focus on the tools that actually run them . We'll lay out the philosophical difference between PyTorch and TensorFlow, and trace how Hugging Face transformers , llama.cpp , MLX , and Ollama built the bridge to running large language models on your own machine. By the end you should have the full mental model of "download a pretrained LLM and serve it locally." 0. Learning Objectives Explain the eager-vs-graph difference between PyTorch and TensorFlow. Explain, in graph terms, how autograd automates the backward pass. Use Hugging Face transformers ' AutoModel , AutoTokenizer , and pipeline abstractions. Describe llama.cpp's GGUF quantization, INT4 inference, and CPU-first flow. Describe Apple MLX's use of unified memory and how it differs from PyTorch. Run a local LLM with Ollama and call it through an OpenAI-compatible API. ...
Read more »

"RAG Core Study (14/26) — Evaluation Sets with RAGAS & DeepEval"

이미지
Series overview: Series index 📚 Series Index Without an evaluation set, every RAG improvement is a story, not evidence. RAG systems fail in more than one place: retrieval, context selection, answer generation, and grounding. That means "looks better" is not a metric. Part 14 explains how to build a golden dataset , how RAGAS and DeepEval fit into the workflow, and why teams must separate retrieval quality from answer quality if they want tuning decisions to hold up. 0. Prerequisites Part 13 reranking — system quality now depends on multi-stage ranking. Part 1 grounding — a correct-looking answer is not enough. Part 12 Hybrid — multiple retrieval settings need comparison on fixed data. 1. Learning Objectives Build a small but useful golden dataset for RAG. Distinguish retrieval metrics from answer metrics . Use RAGAS and DeepEval in the right roles. Avoid the common traps in judge-model-based evaluation. 2. 핵심 요약 An evaluation set for RAG is ...
Read more »

"ML Foundations (8/9) — Deep Learning Architectures: CNN, RNN, Attention"

이미지
← 7/9 Deep Learning Training 📚 Series Index 9/9 PyTorch vs TensorFlow → What did we keep adding on top of an MLP? Vision went CNN, sequences went RNN/LSTM, and eventually both converged on Attention and the Transformer. Each architecture is easier to remember if you read it as what it gave up to gain something else . This part is the one-line summary of the decisive inflection points: LeNet → AlexNet → ResNet → LSTM → Attention → Transformer. 0. Learning Objectives Explain why CNN's convolution, pooling, and weight sharing match images so well. Trace what got unblocked from LeNet → AlexNet → VGG → ResNet as depth grew. State BPTT for RNNs and the vanishing/exploding gradient problem. Write the LSTM gate equations (forget, input, output) with intuition. Write the attention formula in query/key/value form and the scaled dot-product variant. State the precise reasons Transformers replaced RNNs (parallelism + long-range dependencies). 1. 핵심 요약 CNN : weight sharin...
Read more »

"ML Foundations (7/9) — Deep Learning Training: Optimizers, Regularization, Initialization"

← 6/9 Neural Networks 📚 Series Index 8/9 Deep Learning Architec… → Building an MLP, as we did in Part 6, is not the same as getting it to train well . The right optimizer, regularization, initialization, and learning rate — when these line up, deep networks converge. When they don't, the network refuses to learn at all. This part is the catalog of those crafts, with formulas, paper citations, and code in one place. 0. Learning Objectives Compare and write the update rules for SGD, Momentum, Nesterov, and Adam. Explain how Dropout, BatchNorm, and LayerNorm work and where they belong in a model. Derive the variance formulas for Xavier and He initialization and match them to activations. Implement step, cosine, and warmup learning-rate schedules in PyTorch. Explain why gradient clipping is effectively required for RNNs and Transformers. Diagnose the most common training failures (NaN, plateau, overfitting) and apply first-line fixes. 1. 핵심 요약 SGD : \(w \leftarro...
Read more »

OpenClaw to Hermes Migration (2/13) — What to Preserve, Partially Port, or Discard

← 1/13 Current Structure Inve… 📚 Series Index 3/13 What Moves to Hermes → A code review record classifying operational scripts into three tiers: preserve, partial-preserve, and discard. What This Post Covers An asset classification framework (preserve / partial-preserve / discard) for migrating a legacy agent system to a successor A concrete case applying the Strangler Fig + Subset Migration strategy to a memory pipeline Classification results evaluated across LOC, external dependencies, and replaceability axes A method for slimming down a bloated single file (1,310 LOC) using the migration as a natural entry point Problem Definition: Port Everything or Cut Deep? The OpenClaw inventory comprises multiple agents, multiple scheduler/daemon processes, and multiple operational scripts. When moving to the successor system, Hermes, the choice is not binary. A full port carries accumulated technical debt along with it; a minimal port discards hard-won domain logic. Code-revie...
Read more »

AI Agents I Built (5/7) — Building an Automated Blogger API Publishing System

← 4/7 My Life Organizer Agent 📚 Series Index 6/7 Dual → Designing a Markdown-Based Auto-Publish Pipeline with Blogger API v3 + OAuth 2.0 Key Summary Blogger API v3 + OAuth 2.0 enables a pipeline that converts Markdown files into publishable HTML posts blogger_publish.py handles the full flow — frontmatter parsing → body cleanup → HTML conversion → CSS injection → API post — in a single script For bulk publishing, quota management (delay, retry, TOC suppression) is the critical design concern Platform Selection Criteria API access is mandatory for AI agent-driven content publishing. A comparison of platforms: Platform API Cost AdSense Decision WordPress.com REST API Paid plan required Paid only Cost overhead Ghost REST API Self-hosted or paid Manual setup Ops overhead Blogger v3 API Free Native integration Adopted Blogger is free, exposes a REST API, and integrates natively with AdSense. For personal blog automation, this combination i...
Read more »