Skip to main content
used mini pcs

Your own Zapier

Self-hosted n8n and workflow automation on a used mini PC

n8n, Node-RED, Huginn and Activepieces can replace your Zapier-style subscriptions with a server you own, and none of them needs much hardware. We'd run them on a used 1-liter business PC. It runs 24/7 at low idle power and has room left for the database, backups and a tunnel for webhooks.

What you need

SpecRecommendedWhy
CPU generationIntel 6th gen or newer, Intel N-series, or any Ryzen; 4 cores is plentyn8n's own benchmark ran a single instance (main mode, Postgres) on an AWS c5a.large with 4 GB of RAM and handled up to 220 workflow executions per second. That's far beyond a home or small-business workload.
RAM8 GB minimum, 16 GB comfortable, 32 GB if Postgres, Redis and other services share the boxn8n's docs say memory use depends on how much JSON and binary data a workflow handles, how many run at once, and whether you use the Code node or manual executions, which copy data for the editor. Big payloads crash small n8n servers long before the CPU runs out.
Storage typeNVMe or SATA SSD, 256 GB or largern8n keeps workflows, credentials and execution history in SQLite by default, or in PostgreSQL. Databases do many small writes, which is SSD work, and execution history and binary files grow over time.
NICsOne wired 1 GbE portWorkflows are API calls and webhooks, which are small, latency-sensitive requests with little bulk transfer. Wired beats Wi-Fi for uptime.
Idle wattsAbout 12 to 15 W at the wall for a 1-liter business tinyA used ThinkCentre M920q measured 12 to 15 W idle. An automation server waits for triggers almost all day, so idle draw is your running cost.
NoiseQuiet at idle; brief fan spin-ups under heavy jobsThe M920q is barely audible at idle, and its fan only picks up under sustained CPU load, which workflow servers rarely produce.
Remote managementNice to have: Intel vPro (AMT)On a headless box in a closet, AMT lets you power-cycle it and reach the console remotely. vPro-configured M920q units have AMT 12.0.

Which mini PCs fit

A model gets the “n8n / automation” badge when it:

  • Takes at least 8 GB of RAM (manufacturer's stated maximum)
  • Has an M.2 slot that takes an NVMe SSD
  • Has wired Ethernet

We only use the makers’ specs. 47 of the 58 models we cover qualify.

Best overall

Lenovo ThinkCentre M920q Tiny

A box tucked away in a closet wants remote management, and vPro configurations of this tiny have Intel AMT 12.0 on an I219-LM NIC. It takes 8th/9th-gen T-series CPUs up to 8 cores and an NVMe drive plus a 2.5-inch one. Lenovo says 32 GB, but ServeTheHome had 2 × 32 GB SO-DIMMs working, and it measured 12 to 15 W idle.

On eBay now: A quad-core i3-8100T with an SSD and Windows 11 Pro for $137 delivered. Four cores is plenty for Home Assistant, and ask which adapter is included. $136.84

View on eBay ↗: Lenovo ThinkCentre M920q i3-8100T 3.1GHz 8GB Ram 256GB SSD WINDOWS 11 PRO

Best value

Lenovo ThinkCentre M720q Tiny

Same chassis family as the M920q, minus vPro, and far more common used. It has 8th/9th-gen CPUs, two DDR4-2666 SO-DIMM slots (32 GB official), NVMe plus a 2.5-inch bay, and an Intel I219-V NIC. A quad-core i3 or six-core i5 unit with 16 GB runs n8n, Node-RED and Postgres together without strain.

On eBay now: A quad-core i3-8100T with an M.2 SSD for $115 delivered. Fine for Home Assistant or Pi-hole, but ask about the adapter. $115.00

View on eBay ↗: Lenovo ThinkCentre M720q Intel Core i3 8100t 8GB 256GB SSD M2

What it costs to leave on all year

An always-on box spends almost all its time near idle, so idle draw sets the bill. Here’s the sum.

idle watts × 24 h × 365 days ÷ 1,000 = kWh per year · kWh × $0.1831/kWh = dollars per year

MachineIdlekWh / yearCost / year
1-liter business tiny (used Lenovo ThinkCentre M920q), idle at the wall, measured by ServeTheHome, low end source12 W105.1$19.25
1-liter business tiny (used Lenovo ThinkCentre M920q), idle at the wall, measured by ServeTheHome, high end source15 W131.4$24.06
Beelink SER5 (Ryzen 5 5500U / 5600H / Ryzen 7 5700U) (measured, 5.3 W upper figure) source5.3 W46.4$8.50
Minisforum UM790 Pro (Ryzen 9 7940HS) (measured, 7 W upper figure) source7 W61.3$11.23
Intel NUC 13 Pro (NUC13ANK / NUC13ANH) (measured, 8 W upper figure) source8 W70.1$12.83
Lenovo ThinkCentre M75n Nano (measured, 8 W upper figure) source8 W70.1$12.83
Beelink SER6 Pro (Ryzen 7 6800H / 7735HS) (measured, 8.2 W upper figure) source8.2 W71.8$13.15
ASUS NUC 14 Pro (measured, 9 W upper figure) source9 W78.8$14.44
Beelink GTR6 (Ryzen 9 6900HX) (measured, 10 W upper figure) source10 W87.6$16.04
Dell OptiPlex 3070 Micro (measured, 10 W upper figure) source10 W87.6$16.04

Electricity at 18.31¢/kWh, the U.S. average residential price, July 2026 (EIA). Your own rate is on your bill, so swap it into the sum. Real draw rises above idle whenever the box is working, and it depends on the drives, RAM, BIOS power settings and OS.

What n8n actually needs

n8n doesn't publish a minimum hardware spec for self-hosting, but its docs tell you enough to size a box. Its benchmark page reports up to 220 workflow executions per second from a single n8n instance in main mode with a Postgres database, on an AWS c5a.large with 4 GB of RAM. A home lab or small team runs a tiny fraction of that, so CPU is almost never what holds you back.

Memory is. n8n's guide to memory problems lists what drives usage. That's the amount of JSON data, the size of binary data, the number of nodes in a workflow, how many executions run at once, the Code node (which it says can increase memory use significantly), and manual executions, which copy the data for the editor. Its fixes are to split data into smaller chunks, use sub-workflows, and raise Node's heap limit with `--max-old-space-size` through `NODE_OPTIONS`. In practice, 8 GB of RAM is enough for n8n alone, 16 GB is comfortable, and 32 GB lets the same box carry Postgres, Redis and other containers.

Use Docker, then pick SQLite or Postgres

Install n8n with Docker. The documented command maps port 5678 and keeps data in an `n8n_data` volume mounted at `/home/node/.n8n`. The npm route requires Node.js 20.19 to 24.x, is deprecated from n8n 3.0, and n8n's own docs call it not safe for production. n8n is also candid about the job. It says self-hosting takes technical knowledge, including setting up servers and containers, managing resources and securing the server, and that mistakes can lead to data loss, security issues and downtime. Plan backups from day one.

By default n8n stores credentials, workflows and execution history in SQLite. It also supports PostgreSQL, currently versions 16, 17 and 18. SQLite is fine for a single-user box. Postgres is the better long-term home if you'll keep a lot of execution history or scale out later. Either way, put the database on an SSD.

Node-RED, Huginn and Activepieces

  • Node-RED is flow-based programming that runs on Node.js. Node-RED 5.x needs Node.js 22 or later, 4.x needs Node.js 18, and the project currently recommends Node 24. It's light enough to share a box with everything else here, and it also installs as an app inside Home Assistant OS.
  • Huginn calls itself a hackable version of IFTTT or Zapier on your own server, with agents that monitor and act on your behalf. The quickest route is its official Docker image.
  • Activepieces calls itself an open source replacement for Zapier. The Community Edition is MIT-licensed, and the recommended self-hosted production setup is Docker Compose with PostgreSQL and Redis, which is a good reason to buy a box that takes 32 GB.

Cron, systemd timers and plain scripts

Not every automation needs a platform. A shell or Python script on a systemd timer or cron schedule is the lightest automation there is, and an always-on mini PC is the natural place to run it. Think nightly backups, scraping a price, renaming downloads or pinging a health check. When a script grows branches, retries and credentials, move it into n8n or Node-RED. If the automation needs an AI agent that writes code or browses the web, see the always-on AI agent server page. For a local model inside workflows, see local LLMs on a mini PC.

Webhooks without opening your router

Many workflows start from a webhook, so something on the internet has to reach your box. Don't forward a port. Cloudflare Tunnel's `cloudflared` connects outbound to Cloudflare, so you need no publicly routable IP and can block all inbound traffic. Tailscale Funnel exposes a local service to the internet on port 443, 8443 or 10000 over TLS, and it's available on all Tailscale plans. Keep n8n's editor UI private (LAN or Tailscale only) and expose only the webhook path.

Mini PC vs cloud hosting

A small VPS is the alternative, and it wins on public uptime and a static IP. A mini PC wins on monthly cost, since there isn't one beyond electricity, and it keeps the credentials for your email, CRM and bank-feed integrations at home. At the 12 to 15 W an M920q draws at idle, the yearly electricity estimate on this page stays small. If you also want to run network services or a Proxmox homelab, one 32 GB tiny can host all of it.

ThinkCentre vs OptiPlex vs EliteDesk compares the big three business brands, and the buying checklist covers what to check on any used unit.

Setup outline

  1. 1Before money changes hands, ask if a BIOS supervisor (administrator) password is set. It blocks boot order, UEFI and Secure Boot changes, and if the seller can't remove it, buy a different unit.
  2. 2Check BIOS security for an activated Computrace or Absolute setting on first boot. Absolute's module lives in firmware on many brands and reinstalls its agent after a wipe, so an activated one means asking for proof the previous owner released the device.
  3. 3Look up the machine by serial number on the maker's support site and install the newest BIOS. Old firmware usually explains odd memory or NVMe behavior on off-lease units.
  4. 4Install RAM as a matched pair (16 GB or more) and an SSD of 256 GB or more, since the databases and execution history live there.
  5. 5Run a full pass of MemTest86+ before you trust the box with credentials and scheduled jobs.
  6. 6In BIOS, set power-on after AC loss, then install Ubuntu Server 24.04 LTS and Docker with the Compose plugin.
  7. 7Start n8n with the official Docker image and a persistent `n8n_data` volume on port 5678. Add PostgreSQL in the same Compose file if you expect lots of execution history.
  8. 8Expose only the webhook endpoints, through Cloudflare Tunnel or Tailscale Funnel, and keep the editor private. Schedule nightly backups of the n8n data volume and database to another disk or machine.

n8n / automation: common questions

What hardware do I need to self-host n8n?

Very little CPU. n8n's benchmark handled up to 220 executions per second on a single 4 GB AWS c5a.large instance. Memory and storage matter more, so plan on 8 GB of RAM minimum, 16 GB to be comfortable, an SSD for the database and a wired network port. Any used 6th-gen-or-newer 1-liter tiny qualifies.

Should n8n use SQLite or PostgreSQL?

SQLite is n8n's default and is fine for a single user. n8n also supports PostgreSQL (currently 16, 17 and 18), and that's the better choice if you'll keep a lot of execution history or scale out later.

Why does my n8n server run out of memory?

n8n's docs list the causes, which are large JSON or binary payloads, many nodes, simultaneous executions, the Code node, and manual executions, which copy data for the editor. Split data into chunks, use sub-workflows, raise the Node heap limit with NODE_OPTIONS, or add RAM.

What's the best self-hosted Zapier alternative?

It depends on how you like to work. n8n is the most popular visual workflow tool. Node-RED is flow-based and strong for device and home-automation work. Activepieces calls itself an open source replacement for Zapier (MIT-licensed Community Edition), and Huginn is a hackable, agent-style option. All four run in Docker on the same mini PC.

Can I run n8n and Home Assistant on the same mini PC?

Yes. Run them as Docker containers next to a Home Assistant VM under Proxmox, or install Node-RED as an app inside Home Assistant OS. A tiny that takes 32 GB gives both plenty of room.

How do webhooks reach n8n at home?

Use Cloudflare Tunnel or Tailscale Funnel instead of port forwarding. Both make an outbound connection, so nothing on your router has to be opened, and you can expose only the webhook path while the editor stays private.

Sources

Run more on the same box

Upgrading? Sell the old one to us.

Tell us what you have (model, CPU, RAM, drive, whether there’s a BIOS password), add a few photos and your price, and we’ll reply with an offer by email. No listing, no fees.

Get an offer