--- title: "Run n8n Without Docker: 3 Setup Routes Compared" description: "npm install vs cloud hosting vs one-click: three realistic ways to run n8n without Docker, compared on effort, maintenance, and AI-node readiness." canonical: https://agentroost.app/en/blog/run-n8n-without-docker-routes-compared date: 2026-05-18T12:00:00Z --- [Canonical URL](https://agentroost.app/en/blog/run-n8n-without-docker-routes-compared) Most people who Google "run n8n without Docker" have already read the official docs, noticed the install command starts with `docker run`, and closed the tab. Fair. Docker is a reasonable tool for developers who use it daily, but if you're a marketer, ops manager, or solo founder who just wants n8n's workflow editor running — Docker is a yak shave, not a feature. The good news: Docker is genuinely optional. The bad news: the alternatives each come with their own caveats. This post walks you through all three realistic routes so you can pick one that actually fits your situation. --- ## The Three Routes ### Route 1: `npm install n8n` — Local Machine n8n ships as a Node.js package. If you have Node 18+ installed, this works: ```bash npm install -g n8n n8n start ``` Your editor opens at `http://localhost:5678`. No Docker, no container runtime, nothing. Workflows you build persist in `~/.n8n/` as SQLite by default. **What it's good for:** Testing n8n locally before committing to anything. Building a workflow, exporting it, understanding the node library. **What breaks in practice:** - It only runs while your laptop is open. A Schedule Trigger node that fires every day at 9 AM silently does nothing when your machine is sleeping. - Webhooks (`/webhook/…` URLs) point to `localhost`. External services like Stripe, GitHub, or Typeform can't reach them. You'd need a tunnel (ngrok, Cloudflare Tunnel) to receive anything from the internet — and those tunnels die when your laptop restarts. - Upgrades are manual: `npm update -g n8n`, then cross your fingers nothing broke. - AI/LLM nodes need OpenAI, Anthropic, or another API key you supply yourself. The nodes appear but produce errors until you wire up credentials. **Verdict:** Good for exploration. Not viable for anything you actually want to *run*. --- ### Route 2: VPS + `npm` (or Docker Compose) — Your Own Server Rent a $6/mo VPS (Hetzner, DigitalOcean, Fly.io), SSH in, install Node, run n8n as a process managed by `pm2` or `systemd`. Or use Docker Compose if you're willing to learn just enough Docker to paste a `docker-compose.yml`. With a VPS you get: - A real public IP — webhooks work, Schedule Triggers fire 24/7. - Full control over your data and config. - Freedom to run Postgres instead of SQLite for reliability. But you also take on: | Task | Who does it | How often | |---|---|---| | SSL certificate (HTTPS) | You (Caddy/nginx + Let's Encrypt) | Initial + renewal | | OS security patches | You | Ongoing | | n8n version upgrades | You | Every release | | Postgres backup | You | Daily ideally | | Port firewall rules | You | Initial + any change | | PM2/systemd restart on crash | You | Configure once, debug when it stops working | None of these are *hard*, but each one is a context switch away from the workflow you actually wanted to build. And when something breaks at 2 AM because a dependency updated, you're the on-call engineer. **AI nodes on a self-hosted VPS:** You still need your own API keys. The OpenAI node, Anthropic node, or any AI/LLM node in n8n asks for credentials you paste in. That's extra cost on top of your VPS bill, tracked and billed separately by each provider. **Verdict:** The right call for developers who want full control and don't mind the ops overhead. Not the right call if "ops" is not your job. --- ### Route 3: One-Click Hosting — Your Own Instance, No DevOps A third category has emerged: providers that spin up a pre-configured n8n instance for you, give you a public URL, and handle all the infrastructure underneath. You get the full n8n editor — your login, your workflows, your data — without touching a server. This is distinct from a shared hosting environment: **you own the instance**. Your workflows run in isolation on hardware dedicated to your workspace, and your data doesn't commingle with other users' data. It's self-hosting without the self-hosting. What you avoid: - Choosing a VPS provider and region - Installing and configuring Node, Postgres, nginx, Certbot - Writing a `docker-compose.yml` (or learning Docker at all) - Setting up PM2/systemd and figuring out why it's not restarting - Renewing SSL certs - Upgrading n8n between versions **The AI-nodes question** is where one-click hosting splits into two camps: | Provider | Hosting model | AI nodes need | |---|---|---| | Elestio | Docker-based deploy | Your own API keys | | Sliplane | Docker-based deploy | Your own API keys | | n8n Cloud (official) | Hosted by the n8n team | Your own API keys | | **AgentRoost** | **Your own isolated instance** | **AI credits included — no keys** | Every competitor in this space is bring-your-own-API-key (BYOK). You still have to sign up for OpenAI, paste the key into n8n's credentials, and manage a separate bill. On **AgentRoost**, LLM/AI credits are bundled into the subscription price. The AI nodes — the ones that call an LLM, generate text, classify data — are pre-wired and working the moment your instance starts. You don't configure anything. --- ## Honest Comparison Table | | Local npm | VPS (self-managed) | One-click hosting | |---|---|---|---| | Docker required? | No | Optional | No | | Works 24/7? | No | Yes | Yes | | Webhooks (public HTTPS)? | No (tunnel needed) | Yes (you set up SSL) | Yes (auto-provisioned) | | AI nodes work out of the box? | No | No | Yes (on AgentRoost) | | Ops burden | Low start, breaks silently | High | None | | Upgrade n8n | Manual | Manual | Managed | | Monthly cost | ~$0 | $6–12 VPS + API keys | From $19.99 all-in | | Data control | Local only | Full | Your isolated instance | --- ## Who Should Pick What **Pick local npm if** you want to learn n8n's node library before committing to anything. Build a workflow, export it as JSON, understand what nodes you need. Treat it as a sandbox, not a production setup. **Pick a self-managed VPS if** you're a developer who wants root access, plans to run other services on the same machine, and is comfortable with Linux administration. Budget an hour upfront and occasional maintenance windows. **Pick one-click hosting if** you want a working n8n instance — with AI nodes that actually fire — without the ops overhead. Especially if you're building automations that involve LLM calls, AI classification, or text generation and you don't want to track a separate API key bill. --- ## How to Do It on AgentRoost (the No-Docker Route) 1. Sign up at agentroost.app (email/password, Google, Microsoft, or Discord). 2. From your dashboard, pick **n8n** as your framework. 3. Name your instance — your editor will open at `https://.agentroost.app`. 4. Open any workflow, drop in an AI/LLM node (e.g., the **OpenAI** node or the built-in **AI Agent** node), and run it. The credentials are already wired to your included credits. No API key prompt. 5. Build your workflow. Webhooks get a real public HTTPS URL automatically — paste it into Stripe, GitHub, Typeform, or whatever triggers your flow. Pricing starts at **$19.99/mo** — that's roughly what you'd spend on a $6 VPS plus a separate LLM API credit budget, bundled with zero setup and a 14-day money-back guarantee. Plus and Pro tiers add more compute and more included AI credits if your workflows are heavier. [Compare plans](/en/pricing) — or go straight to [the n8n instance page](/en/agents/n8n) to see what's included. --- ## A Note on AI Nodes Specifically If your use case is purely data transformation — copying rows from a spreadsheet, sending Slack messages, chaining HTTP requests — any of the three routes will serve you. The n8n core is open-source and the non-AI nodes work identically everywhere. But if you're building anything that involves: - Summarizing emails or documents - Classifying customer feedback - Generating draft replies or content - Running an AI agent that decides which branch of a workflow to take - Calling a local or hosted LLM to process data ...then the BYOK tax starts to matter. On a self-managed setup you're paying for the VPS *and* the API credits *and* your time. On n8n Cloud you're paying a hosting fee *and* the API credits. On AgentRoost the credits are folded in — you pay one price and the AI nodes just work, across 350+ LLM models you can switch between at any time. --- The "run n8n without Docker" question usually means "I want n8n without the complexity." Local npm gets you there for five minutes; a one-click instance gets you there permanently.