// blog

We Docker-Deployed Every n8n Alternative. The Results Don't Match the Listicles.

Search “n8n alternatives” and you’ll find the same article rewritten a dozen times: a feature table copied from vendor docs, a screenshot of each homepage, and a verdict that reads like it was written by whoever answered the sponsorship email first. What none of them tell you is the thing that actually decides whether a tool works on your server: what happens when you run docker compose up.

So we ran it. All six of the most-recommended self-hosted automation tools, one after another, on the same machine, with the same procedure: write a compose file from official docs, deploy, wait for HTTP 200, record idle RAM, and write down everything that went wrong. The results cut against several pieces of conventional listicle wisdom — including which “easiest n8n alternative” actually deploys easiest.

Tested on: Ubuntu 26.04 LTS, Xeon W-2155, 128 GB RAM, Docker CE 29.7.2 / Compose v5.5.0. Cold-pull time excluded except where noted; every stack was measured to its first working page (HTTP 200). Versions were latest images as of 2026-08-23.

The matrix

Idle numbers are steady-state after boot, measured with docker stats. Full raw notes are in the methodology section below.

What the numbers actually say

Activepieces is the hardest “easy alternative” to self-host

This was our biggest surprise. Activepieces markets itself as the open-source Zapier alternative, and listicles consistently rank it the friendliest n8n swap. On our box it took three config iterations to get running:

1. First attempt with SQLite config failed at boot — current releases no longer support SQLite.
2. Adding Redis wasn’t enough; the app crash-looped demanding AP_POSTGRES_DATABASE and friends.
3. Only the full Postgres + Redis + app stack came up healthy.

Final footprint: three containers, roughly 914 MiB combined, with the app itself the single hungriest process we measured (724 MiB idle). None of this means you shouldn’t use it — the UX is genuinely good — but if your mental model of self-hosting is “one container and go,” Activepieces is the furthest thing from it among these six.

Node-RED is the featherweight — with an asterisk

One container, 91 MiB idle, serving pages in under eight seconds, zero configuration issues. It’s four times lighter than anything else here. The asterisk: Node-RED isn’t trying to be Zapier. It’s a flow-based wiring tool with an IoT heritage and a node palette rather than an integration app catalog. If your automations are device-and-API plumbing, nothing else in this test comes close on efficiency. If you want prebuilt connectors for SaaS apps with OAuth flows handled for you, the lightness won’t matter — you’ll spend the saved RAM building what other tools ship.

n8n hits the deployment sweet spot

The baseline everyone compares against is also the best-balanced: a single container with embedded SQLite, 356 MiB idle, no external database to babysit. Our only issue was self-inflicted-adjacent — a bind-mounted data directory created by root caused an EACCES crash loop until we chowned it to UID 1000 (the container’s node user). That’s a five-second fix once you know it, but it’s exactly the kind of thing that produces a GitHub issue titled “container exits immediately.”

Worth noting the license column too: n8n’s Sustainable Use license is fair-code, not OSI open source. For personal and internal business use it changes nothing; if you’re embedding it into a product you sell, read it before you build.

Windmill: the best-engineered deploy, aimed at developers

Windmill runs the biggest stack — Postgres plus separate server and worker containers, ~1.46 GiB all-in — and it was the only multi-container tool that worked on the very first compose up. No missing environment variables, no crash loops, no permission surprises. That’s not luck; the official compose file is complete.

It’s also the least Zapier-like tool here besides Node-RED. Windmill treats workflows as code (TypeScript, Python, Go) with a UI on top. If you think in scripts and version control, it may be the best fit on this list. If you think in drag-and-drop nodes, it won’t.

Automatisch: honest Apache-2.0 licensing, two small traps

Four containers (app, worker, Postgres, Redis) but a modest ~259 MiB total. Two things bit us:

• The image lives at automatischio/automatisch on Docker Hub — automatisch/automatisch does not exist, and the error message doesn’t hint at it.
• ENCRYPTION_KEY and WEBHOOK_SECRET_KEY aren’t optional. Without them the app crash-loops during database migration with a generic knex error.

Both are documented if you read carefully; both are easy to miss. Automatisch remains notable as the fully-open-source option (Apache-2.0, no fair-code caveats), though its connector catalog is smaller than n8n’s or Activepieces’.

Huginn: alive, but showing its age

The oldest project here still works. Bundled MySQL 8, ~1.2 GiB total, and a first-boot race where the Rails app started before MySQL finished initializing — one manual restart fixed it. Huginn’s agent model (scheduled, composable agents rather than flow canvases) remains genuinely distinctive for data-scraping style automations. But nothing about the deployment experience has modernized, and the RAM cost is second-highest per unit of polish.

Methodology and reproducibility

Every tool got the same treatment on the same box, in order: n8n → Activepieces → Windmill → Node-RED → Automatisch → Huginn.

• Compose files written from each project’s official docs, minimal config, localhost-only ports.
• Deployment = docker compose create && docker compose start, then poll the mapped port for HTTP 200.
• Idle RAM via docker stats --no-stream after boot settled.
• Gotchas recorded verbatim from container logs at failure time.
• One caveat we’ll own: our timings reflect warm image pulls for five of six stacks (n8n’s includes its pull). Relative difficulty, not absolute seconds, is the transferable finding.

The full results table, ports, and raw gotchas live in our repo notes; every claim above traces to a logged container error or a docker stats line.

Which one should you actually run?

• Want one container and zero databases? n8n (356 MiB) — the default for a reason.
• Running on a Raspberry Pi-class box? Node-RED (91 MiB), if flow-style automation fits your use case.
• Want true OSI open-source licensing? Automatisch (Apache-2.0) or Activepieces core (MIT).
• Think in code, want workflows in git? Windmill — and its deploy is the cleanest here despite the size.
• Scraping/scheduled-agent workloads, nostalgia included? Huginn still functions.
• Following a listicle that calls Activepieces the easiest switch? Now you have numbers that say otherwise.

Self-hosted automation tools differ less on features than listicles imply and more on operational shape: how many processes you run, how much RAM they drink at idle, and how many times compose up fails before it works. Those are the columns nobody publishes — so we did.