n8n Alternatives (2026): What to Switch To, and When Not To
Most n8n alternatives lists are written by a vendor selling one of them. Verified licences and prices, plus the reasons to switch that self-hosting fixes.
By Ali Ilyas · · Updated September 16, 2026
Search for n8n alternatives and you get lists published by Composio, ZenML, Gumloop and Vellum. Each one sells an automation or agent product, and each one appears in its own list. The recommendations are not necessarily wrong, but you are reading a sales page with a comparison table on it, and nobody writes "stay where you are" on a page designed to move you.
This page sells nothing and carries no affiliate links. What follows is every tool's licence read from its own repository, prices read from its own pricing page, both on 2026-09-16, and an argument that three of the five common reasons for leaving n8n are solved without leaving.
What's on this page
- Which of the five reasons is yours?
- Three reasons that do not need a different tool
- The reason that is real: the licence
- The alternatives, with licences checked today
- If you want open source in the strict sense
- If you want somebody else to run it
- If you would rather write code than wire nodes
- What you give up by leaving
- Migration: what has to be rebuilt
- How I would choose
- Frequently asked questions
Which of the five reasons is yours?
People arrive at this search with one of five complaints. The right answer depends entirely on which one, and two of them point at a different tool while three point back at your own setup.
- n8n Cloud costs more than you want to pay.
- The integration you need does not exist.
- It gets slow, executions queue up, or the instance falls over.
- You want to resell or embed it, and the licence says no.
- You do not want a canvas at all, because your team writes code.
Reasons 4 and 5 are genuine grounds for switching. Reasons 1 through 3 are configuration problems wearing a product-choice costume, and swapping tools will reproduce all three under a different logo.
Three reasons that do not need a different tool
The bill
n8n Cloud starts at $20/month billed annually for 2,500 executions, then $50 for 10,000 and $800 for 40,000 (checked 2026-09-16). Self-hosted Community edition has no execution cap and costs whatever your server costs, usually $5 to $8 a month.
Before you compare n8n against Make, compare n8n Cloud against n8n on a VPS. We broke down both sides in what n8n actually costs, and the Docker setup is a twenty-minute job. Most people who leave over price were paying for hosting, not for software.
The missing integration
n8n ships 274 documented app nodes and 90 trigger nodes, which still leaves gaps. The gap is usually filled by the HTTP Request node rather than by a different platform, because nearly every service worth automating has a REST API and n8n's credential system handles the auth around it. If the vendor has an API and you switch tools instead, you have traded a twenty-minute HTTP node for a migration.
The performance
Slowness at volume is nearly always the default single-process setup doing work that belongs in queue mode, or a workflow pulling an entire dataset into memory when it could page through it. Executions that hang or never fire have separate causes, most of them fixable: when a workflow will not trigger covers the usual suspects. A tool that is fast on your first test workflow will be slow on your hundredth too, for the same reasons.
That leaves the two complaints that survive contact with a fix.
The reason that is real: the licence
n8n is not open source, whatever the GitHub repository looks like at a glance. It ships under the Sustainable Use License, and the limitation is a single paragraph:
"You may use or modify the software only for your own internal business purposes or for non-commercial or personal use. You may distribute the software or provide it to others only if you do so free of charge for non-commercial purposes."
That is from LICENSE.md in the n8n repository, read on 2026-09-16. Files with .ee. in the name or .ee in the directory path sit outside even those terms and require an enterprise licence.
Running client automations on your own instance is fine. Building a product where your customers log into n8n, or bundling it into something you sell, is not. We work through the boundary cases in n8n commercial use, including the two conditions people miss.
If your business model needs the software itself to be freely redistributable, no amount of self-hosting fixes that. This is the one complaint on the list where switching is the answer rather than a detour.
The alternatives, with licences checked today
Licences come from each project's own LICENSE file and the activity dates from the GitHub API, both read on 2026-09-16. Stars are a popularity signal and nothing more.
| Tool | Licence | Self-host | Stars | Last push | Shape |
|---|---|---|---|---|---|
| n8n | Sustainable Use (fair-code) | Yes | 204.5k | 2026-09-16 | Visual canvas, 274 app nodes |
| Activepieces | MIT core, packages/ee/ separate | Yes | 24.5k | 2026-09-16 | Visual canvas, closest in feel to n8n |
| Node-RED | Apache 2.0 | Yes | 23.7k | 2026-09-09 | Visual wiring, event-driven, IoT roots |
| Windmill | AGPLv3, Apache parts, proprietary enterprise | Yes | 18.0k | 2026-09-16 | Scripts first, flows and UIs on top |
| Kestra | Apache 2.0 | Yes | 28.1k | 2026-09-16 | YAML-declared orchestration |
| Automatisch | AGPL-3.0, .ee. files excepted | Yes | 14.0k | 2026-02-11 | Zapier-shaped, see the warning below |
| Huginn | MIT | Yes | 50.0k | 2026-09-16 | Ruby agents that watch and act |
| Temporal | MIT | Yes | 23.1k | 2026-09-16 | Durable execution for code |
| Trigger.dev | Apache 2.0 | Yes | 16.3k | 2026-09-16 | Background jobs and AI workflows in TypeScript |
| Inngest | SSPL v1, Apache 2.0 future licence | Yes | 5.8k | 2026-09-16 | Event-driven step functions |
| Pipedream | Pipedream Source Available v1.0 | Limited | 11.7k | 2026-09-16 | Code-first with a hosted runtime |
| Zapier | Proprietary | No | n/a | n/a | Hosted, per-task billing |
| Make | Proprietary | No | n/a | n/a | Hosted, per-credit billing |
Two rows deserve a second look before you shortlist them. Automatisch is the project most often recommended as "the open source Zapier", and its last push lands on 2026-02-11, seven months before this was written. Pipedream's licence is source-available, which puts it in n8n's category rather than offering an escape from it.
If you want open source in the strict sense
Activepieces
The closest thing to a drop-in. Visual canvas, a piece library the project puts at 760+ integrations, and an MIT licence on everything outside packages/ee/ and the API server's ee directory. That split matters: an MIT core means you can fork, embed and resell it, while the enterprise directory carries its own terms, so read those before you build a product near them.
Cloud pricing on 2026-09-16 runs $0 for 100 credits a day, $20/month flat for Plus, $200/month flat for Team, and embedding from $36k a year. Flat monthly pricing is unusual here and worth noticing if your volume is spiky, because you are not buying execution blocks.
Where it trails n8n: the node library is younger, and the data-shaping work that n8n's expression editor and Code node make routine is less developed.
Node-RED
Apache 2.0, no dual-licence asterisk, and the oldest project on this list by a decade. It grew out of IBM's IoT work, which shows in both directions. Message routing, MQTT and hardware are first-class, the wiring model is genuinely elegant, and the community node ecosystem is enormous. SaaS integrations are where the age shows, and you will write HTTP calls and handle OAuth yourself far more often than in n8n.
Pick it when the work is event plumbing rather than SaaS glue, or when Apache 2.0 with no exceptions is a requirement you cannot negotiate.
Windmill
Built for people who would rather write a Python or TypeScript function and get a schedule, a webhook and a UI around it. The repository describes itself as a developer platform that turns scripts into webhooks, workflows and UIs. Licensing is layered: AGPLv3 on the backend, Apache 2.0 on some parts, proprietary behind the enterprise compile flag. AGPL is itself a real constraint if you plan to offer a modified version as a network service, so it solves n8n's redistribution problem for some business models and not others.
Its cloud pricing is seat-and-worker based rather than execution based, which makes it cheap for a small team running a lot and expensive for a large team running a little.
Kestra
Apache 2.0, and workflows are YAML files rather than canvas state. That one decision makes it the easiest tool here to review in a pull request, diff, and promote between environments. It sits closer to data orchestration than to SaaS automation, so if your workflows look like scheduled pipelines with dependencies and retries, it fits better than n8n does. If they look like "when a form is submitted, message the team", it does not.
Automatisch
AGPL-3.0, Zapier-shaped, and the honest move is to check its commit history before committing to it. The last push recorded by the GitHub API on 2026-09-16 was 2026-02-11. Projects go quiet and come back, but a self-hosted automation tool that stops shipping is a security decision as much as a feature one, because you inherit its dependency updates. Verify the current state yourself rather than trusting this paragraph, or any listicle that placed it second.
If you want somebody else to run it
Neither of these can be self-hosted, so they only make sense if hosting was never the point.
Zapier
Zapier puts its own app count at 9,000+, and it has the shallowest learning curve of anything here. Billing is per task, roughly one per action step that succeeds, and that detail decides your invoice. Free is $0 for 100 tasks a month with two-step Zaps only, so a single condition takes you off it. Professional starts at $19.99/month and Team at $69/month (checked 2026-09-16).
We did this comparison properly in n8n vs Zapier: the real cost math, including where the crossover sits. The short version: per-task and per-execution billing can differ by 10x in either direction depending on how many steps your workflows contain.
Make
Per-credit billing, where one credit is one module action. Free gives 1,000 credits a month with a 15-minute minimum interval between runs. Core is $9/month for 10,000 credits on annual billing, $10.59 paying monthly; Pro is $16 and Teams $29 at the same volume (checked 2026-09-16). Make puts its app count at 3,000+.
Make's canvas handles branching and iteration more comfortably than Zapier's, and its per-credit unit sits between Zapier's per-task and n8n's per-execution. Use the same method as the Zapier comparison: count the actions in one representative workflow, multiply by monthly volume, then compare.
If you would rather write code than wire nodes
This group is not really competing with n8n. It solves the problem underneath it, which is running code reliably over hours or days, and it assumes you have engineers.
Temporal (MIT) is the serious one. Workflows are code, state survives crashes, retries and timeouts are primitives rather than features. It is heavier than n8n by a wide margin, and it is what you reach for when a failed workflow costs real money.
Trigger.dev (Apache 2.0) targets TypeScript teams who want durable background jobs without operating Temporal themselves.
Inngest ships under SSPL v1 with an Apache 2.0 future licence. Read that carefully if redistribution was your reason for leaving n8n, because SSPL is the kind of licence people leave to avoid.
Pipedream is code-first with a hosted runtime and a large connector library, under the Pipedream Source Available License v1.0. Same category as n8n's licence. If reason 4 is what brought you here, this is not an escape.
Huginn (MIT) is the veteran: Ruby agents that watch sources and act on them. No canvas, a steeper setup, a devoted following.
What you give up by leaving
Three things, and the integration count is the smallest of them.
n8n's documentation covers 274 app nodes, 90 trigger nodes and 320 credential types (counted from the docs repository, 2026-09-15 snapshot). Most alternatives are a fraction of that, and the gap shows up on your third integration rather than your first.
The larger loss is the answer supply. Between the community forum, the template library and the volume of published tutorials, something that breaks at 11pm has usually broken for somebody else first. On a smaller project you are reading source instead.
The third thing does not transfer at all: your workflows' failure modes are yours, not n8n's. A workflow that double-writes on retry will double-write on Activepieces. The same goes for the no-rollback problem in agents and for handling bad data mid-run. Migrating is a good moment to fix those and a bad moment to assume they left with the old tool.
Migration: what has to be rebuilt
No importer exists between any two of these tools, and none is coming, because the workflow formats encode different execution models. Plan on rebuilding.
Credentials are the slow part, not the logic. Every OAuth connection has to be re-authorised, which on some services means going back to whoever holds the admin account. Expressions are the other cost: n8n's {{ $json.field }}, along with $fromAI, $node and its item-pairing helpers, has no equivalent on most platforms, so any workflow doing real data shaping gets rewritten rather than transcribed.
Build one workflow on the new tool before you move the other forty, and pick the one with the messiest data handling rather than the simplest. Simple ones port everywhere and tell you nothing.
How I would choose
If price sent you here, self-host n8n and stop shopping. If you need MIT and a canvas, start with Activepieces. If you need Apache 2.0 with no exceptions and the work is event-shaped, Node-RED. If your team writes code and the workflows must not lose state, Temporal, with Trigger.dev as the lighter option for TypeScript shops. If nobody wants to touch a server, Zapier for simple work and Make for branching work, after you do the arithmetic.
And if what you wanted was for n8n to stop falling over, that is a configuration problem. Fixing it is cheaper than migrating.
Frequently asked questions
Is n8n open source?
No, though it is often described that way. The source is public and n8n calls the model fair-code, but the Sustainable Use License limits you to internal business purposes and non-commercial distribution, which the Open Source Definition does not allow. Node-RED, Kestra and Temporal are open source in the strict sense.
What is the best open source n8n alternative?
For the same working style, Activepieces, because the core is MIT and the canvas is familiar. For a licence with no exceptions attached, Node-RED under Apache 2.0. Neither matches n8n's integration count.
Is there a free alternative to n8n?
n8n is already free if you host it yourself, with no execution limit. Among the others, Activepieces, Node-RED, Windmill, Kestra and Huginn are free to self-host. Zapier and Make have free tiers with hard caps: 100 tasks a month on Zapier with two-step workflows only, 1,000 credits a month on Make with a 15-minute minimum interval between runs.
Can I self-host Zapier or Make?
No. Both are hosted products with no self-hosted edition. If running on your own infrastructure is a requirement, they are out regardless of their other merits.
Is Activepieces really MIT licensed?
The core is, under MIT Expat terms. Content under packages/ee/ and the API server's ee directory is licensed separately, exactly as n8n excludes its .ee. files. Read the enterprise terms before building a commercial product that touches those parts.
Which alternative is closest to n8n?
Activepieces, on both the canvas and the mental model. Windmill is closest for teams that want similar coverage but would rather write functions than configure nodes.
Should I switch because of n8n's licence?
Only if you plan to redistribute or embed the software itself. Running automations for paying clients on instances you or they control is permitted, and that covers most freelance and agency work. The licence guide walks through the two questions that decide it.