Gumloop vs n8n: n8n Has 122 AI Nodes. Gumloop Has 15.

Every Gumloop vs n8n comparison I read says the same two things: Gumloop is the AI-native one, n8n is the workflow engine for developers. So I counted the AI nodes in both, straight from their public registries. n8n's AI package registers 122. Gumloop's docs list 15.

That doesn't mean what it looks like, and I'll get to why. But if you're picking one to run your automations for the next year, the agent layer isn't your deciding factor. The license is, and one of these two has a line none of those comparisons quote.

One thing up front. I didn't build a workflow in either.

The short version: which one should you pick?

Pick Gumloop if you want something running this afternoon and never want to see a server. Pick n8n if your automation will run thousands of times or touch a client's data. Gumloop's meter counts every node that fires, so a loop over 100 rows costs 100 times the node price. n8n counts one run of the whole workflow, however many steps sit inside.

How I compared Gumloop and n8n without an account in either

I counted things instead of clicking things, because both tools put building behind a signup. Gumloop's is a normal SaaS account. n8n's is sneakier: the Community Edition really is free and self-hosted, but its first screen asks you to create an owner account with a password. The wall just sits on your own machine.

The inventory is readable without any of that. On July 21, 2026 I pulled two public files. For n8n: curl -s https://raw.githubusercontent.com/n8n-io/n8n/master/packages/@n8n/nodes-langchain/package.json | jq '.n8n.nodes | length'. For Gumloop: curl -s https://docs.gumloop.com/llms.txt | grep -c "docs.gumloop.com/nodes/", which counts node pages in Gumloop's machine-readable docs index. Swap in /nodes/using_ai/ for its AI nodes.

Both take under a minute. Run them before trusting my numbers; these registries move weekly.

n8n: 122 AI nodes on top of 440, and an agent from 2023

n8n is not catching up on AI, and the dates say so. Its AI package, @n8n/nodes-langchain 2.32.0, registers 122 nodes: 24 chat model connectors, 24 vector stores, 12 embeddings, 10 tools, 9 memory nodes and 2 agents. A separate base package adds 440 more for ordinary work.

The Agent node's first commit landed on November 29, 2023, in a pull request titled "Add AI tool building capabilities." AgentHub only renamed itself Gumloop in May 2024. The incumbent shipped its agent six months before the challenger had a name.

Money: the self-hosted Community Edition is free with no execution cap, and n8n's docs say it "includes almost the complete feature set of n8n, except for the features listed here." The exceptions are SSO, projects, environments and git version control. Cloud starts at EUR 20 a month for 2,500 executions, EUR 50 for 10,000, priced in Europe, July 2026.

Who it's wrong for: anyone who wants a finished thing. A chat model connector isn't an assistant, and you assemble the rest.

Gumloop: 15 AI nodes, and that's deliberate

Gumloop documents 300 nodes and only 15 are AI nodes, because its AI nodes are finished jobs rather than building blocks. They read like a to-do list: Ask AI, Categorizer, Scorer, Extract Data, AI Web Research, Generate Report. No vector store to configure, because you're not meant to.

The other 285 are 121 MCP connectors, 114 integrations and 50 utility nodes. The real agent work sits outside that list: Gumloop treats agents as their own product, with skills, memory from company documents, triggers, sessions and evaluations that grade the answers. n8n has no equivalent for that last one. Want to know whether your agent got worse this week? Gumloop has a screen, n8n has your intuition.

Money: free gives you 5,000 credits a month, one seat and one active trigger. Pro is $37 a month for 20,000+ credits and unlimited seats. No card for free, no self-hosting at any price.

Who it's wrong for: anyone whose flow loops over a list. Gumloop's credits page spells it out: "Multiply the node cost by the number of items (e.g. a 60 credit Enrich Contact node over 10 contacts = 600 credits)." That's 12% of your free month on ten contacts.

Gumloop vs n8n: what one run actually costs

n8n charges for the run, Gumloop charges for the run plus every node inside it. n8n's pricing page defines an execution as one run of your entire workflow, step count irrelevant. Gumloop's docs open with "Every workflow costs 1 credit to execute, plus the cost of each node that runs." Its logic, loop and most integration nodes cost 0 credits, so short flows stay cheap. Enrichment, scraping and AI calls are what bite.

The line in n8n's license that no comparison quotes

n8n is not open source, and the difference shows up the moment you do client work. It ships under the Sustainable Use License: "You may use or modify the software only for your own internal business purposes or for non-commercial or personal use." GitHub's API reports its license as "Other", with no open source identifier attached. Every comparison I read called it open source anyway.

For your own automations this changes nothing. Build it once, win every week, on a server you own.

Which one should you pick?

Match the tool to who has to live with the thing, not to which one demoed better.

Solo and non-technical: Gumloop. 5,000 free credits, no card, and its 15 AI nodes are named after the job you want done. Pair it with our one-day AI agent walkthrough and you'll have something running by dinner.

Anything that loops over 50 items: n8n. One execution per run is the cheapest sentence here.

Client work: n8n on their infrastructure, or Gumloop. Avoid your own n8n box hosting five clients' credentials.

An agent you can measure: Gumloop, for the evaluations. Trust me on this one, "is it still working" is the question you'll ask by week three. I turned off half the automations I built in a week.

Would I run either of these tomorrow?

I'd start on Gumloop's free tier and move anything that loops to self-hosted n8n. Boring answer, cheapest one. Those 5,000 credits buy an honest look at whether an automation is worth keeping, and most aren't. The survivors run over a list every morning, exactly where a per-node meter turns ugly.

What I can't tell you is which one you'll enjoy. I counted registries instead of building flows. If you run the same job through both, the number I want is your hundredth run against your first, because neither company puts that on its pricing page. We hit the same wall comparing Lovable and Bolt.