n8n Nod Review: Is This Free Personal AI Agent Worth It?
n8n Nod is a free, open-source personal AI agent from n8n Labs that runs on your own Mac or Windows PC with the AI model you choose. It does what your request covers and is designed to stop and ask before anything else. n8n launched it on November 10, 2026.
The interesting part is the safety story. n8n built real guardrails into Nod, then published a test showing that the AI behind its permission check can still get it wrong. Both halves matter if you plan to let it near your inbox.
Quick note: I've read Nod's docs, privacy pages and code repo, but I haven't run it on my own accounts yet. This is a day-one read of what n8n has published.
What Is n8n Nod?
Nod is a personal assistant you talk to in one chat. It gets its own private computer and browser, walled off from your normal ones, so it can finish errands on real websites.
n8n's examples are small-business chores: chasing an overdue invoice, writing a one-page brief before each client call, cancelling free trials, turning client requests into project tasks, and watching a page for price changes. It connects to Gmail, Calendar, Slack, Notion and Drive, and you can message it from the app, Telegram or WhatsApp.
Nod is its own app. You don't need an n8n account, and it is separate from the n8n AI Assistant inside n8n. The launch post puts it simply: "Nod is for a single player." Nod is for your own errands. n8n stays the tool for shared processes a team relies on.
There is one useful crossover. If you already use n8n, you can connect it and ask Nod to "build an n8n workflow that..." It builds the workflow in your n8n, tests it with sample data, and leaves it as a draft until you say publish. For anyone weighing whether n8n still matters, that is a pretty clear answer from n8n itself.
Nod vs OpenClaw: The Risks We Flagged in March
Back in March, Gael Breton and I explained why you should avoid OpenClaw, the agent that went viral early this year. Our worries were broad access with few checkpoints, prompt injection from pages and emails, exposed keys, and no clear record of what the agent did.
Here is how each of those worries maps to what Nod's docs say it does.
| The risk we flagged | What Nod does | The catch |
|---|---|---|
| Broad access, few checkpoints | A check in Nod's server compares each step that could send, submit or change something with your words and rules. Anything they don't cover waits for your yes. | That check is an AI model, so it can be wrong. Typing into a web field skips it until you press Enter. |
| Prompt injection from pages and emails | Web, app and browser content reaches the model labeled as untrusted data. | The model can still be persuaded. The check then judges whether the action matches what you asked. |
| Exposed keys and passwords | A Vault types passwords, codes and cards into websites privately. Provider keys never enter the workspace or browser. | Nod's server can read a saved value while using it. |
| No record of what it did | A Usage page lists every app action, send, sign-in and card use. | A log helps after a mistake. It doesn't prevent one. |
Some of that is plain software rules. Keys stay out of the browser, passwords get typed privately, and every action gets logged. But the yes-or-no call on each action is made by an AI model, and n8n says so.
And a permission check is a different thing from an accuracy check. Ask Nod to email a client a quote, and the check confirms you asked for that email. It won't confirm the price inside it is right.
The Safety Check Is Only as Good as the Model Behind It
The most important line in Nod's README: "Its safety check is a model's judgement, and Nod is a research preview."
n8n backs that warning with a number. In 240 test decisions, GPT-6.1 Sol and Claude Opus 5.5 got every one right. Claude Haiku 5.5 let 8 unsafe actions through. That's one test set. Treat it as a warning about that model.
So the model behind the check matters. By default, the check runs on the same model you pick for chat, at its lowest reasoning setting. Choose a cheap model to save money, and you've also chosen the model guarding your inbox.
Same care with the headline numbers. Nod reports 98% task completion across 226 errands run three times each, zero unsafe actions in those 678 runs, and 93.2% on a public memory test. Those runs used GPT-6.1 Sol in October 2026, in simulated accounts and websites. The 98% means the task got done. n8n tracks a stricter score separately, where honesty and safety checks also have to pass.
What Nod Shares With n8n
Your files, memory and saved logins stay on the machine running Nod. Your model provider still gets what each task needs, and app connections go through a service n8n hosts, with a partner called Composio, by default.
The part to know: usage data is on by default. It covers which screens and features you use, the kinds of tasks you ask for, timings, costs and failures. It excludes your messages, Nod's replies and what Nod reads. Turning it off doesn't stop task data going to your model provider or the app-connection service.
Conversation traces are off by default. If you turn them on, Nod removes the personal details it recognizes first, but n8n's own example shows a name slipping through.
The wording on this has shifted. An earlier preview copy of Nod's landing page, still online, says: "Beyond update checks, nothing goes to n8n unless you turn on usage data or send feedback." The live page now says n8n "receives usage data to improve Nod, which you can turn off anytime in Settings."
That old line is still out there. When I asked Claude Sonnet 5, with live web search on, "What is Nod by n8n and is it safe to use as a personal AI agent?" on November 11, 2026, its answer quoted the old wording and cited that preview copy.
The privacy page shows a real example of exactly what gets sent. To turn usage data off, go to Settings, then Privacy and data, then Share usage data. On a fresh desktop install, one setup event can send before you reach that switch.
How Much Does n8n Nod Cost?
Nod is free under the Apache 2.0 license, and as of November 11, 2026 there is no paid Nod tier. You pay your AI provider, and that is where the real cost lives.
- ChatGPT Plus or Pro: sign in with your ChatGPT plan, and Nod's usage comes out of the plan's included limits. Other ChatGPT plans fail at sign-in.
- An API key: OpenAI, Anthropic, Google, OpenRouter and 20-plus other providers. You pay per request.
- A local model: no provider bill, but you need a capable computer and some technical setup, and smaller models make the safety check less reliable.
n8n is upfront that costs can add up. The launch post says "some early testers found tasks cost more than they expected." So Nod sets a $200 monthly limit on API spending when you finish setup, and a task checks in with you once it spends $5 since your last message.
Both are Nod's estimates, they only count requests with a known price, and work already running can finish after you hit the limit.
If you already pay for ChatGPT Plus, start there. Usage comes out of a plan you already pay for while you figure out what you'd actually use it for.
Where Nod Falls Short
The biggest limit is practical. Nod only works while the computer running it is on and awake. Close your laptop and the morning brief doesn't arrive. The fix today is an always-on Mac mini or your own server. A hosted version is on n8n's roadmap.
- It's a research preview. n8n calls it an experiment and warns about breaking changes. Version 0.6.2 shipped on November 11.
- One owner per Nod. No team sharing yet.
- Windows is in preview. Running it locally needs 64-bit Windows 11, an extra Windows feature called WSL 2, and 20 GB of free disk space. Macs need macOS 14 or later.
- No phone calls. Calling businesses is listed as upcoming.
Nod's comparison page also has a section called "Where others are ahead." It admits three gaps against Meta's Muse: Nod can't issue single-use card numbers, it has no published bug bounty, and its server can read a saved secret while using it.
Credit to n8n for publishing that. Vendors pick which weaknesses to admit, though, so read it as a reason to keep your main credit card out of Nod for now.
Who Should Try Nod, and Who Should Wait
Nod is worth a test if you run your own business, your errands live in Gmail, Slack and supplier portals, and you'd rather keep your files and logins on your own machine. It's also worth a look if you already use n8n and want an agent that can draft workflows for you.
Wait if you need something that runs while your laptop is closed or that a team can share. The hosted agents in my October best personal AI agent roundup, like ChatGPT dots, Claude Duet and Grok Bot, run on a cloud computer the vendor provides. Nod trades that convenience for control.
My Take on n8n Nod
Nod is the first run-it-yourself agent I've seen designed around the exact risks that made me tell people to skip OpenClaw. Hard rules where they can be hard. A model where judgment is needed. And a public test showing where that model slips.
IMO, here's the line. Let it research, watch pages, prep call briefs and draft replies. Keep client emails, payments and anything you can't undo behind your own yes.
Nod lets you enforce that. Tell it one rule on day one, like "Always ask me before you send email or pay for anything." Per its docs, a rule to ask first wins, even when your request already covers the action.
Let it draft. You hit send.
n8n Nod FAQs
What is Nod by n8n?
Nod is a free, open-source personal AI agent made by n8n Labs and launched on November 10, 2026. It runs on your own Mac, Windows PC or server, uses its own sandboxed computer and browser to finish errands on real websites, and connects to apps like Gmail, Slack and Notion. It asks before any action your request doesn't cover.
Is n8n Nod free?
Yes. Nod is free and open source under the Apache 2.0 license, with no paid tier as of November 2026. You pay your AI provider instead, either through an API key or by signing in with a ChatGPT Plus or Pro plan. With an API key, Nod sets a $200 monthly spending limit at setup and checks in when a task passes $5.
Do I need n8n to use Nod?
No. Nod is a standalone app and doesn't need an n8n account. It is also separate from the n8n AI Assistant. If you do use n8n, you can connect it, and Nod will build, test and fix n8n workflows for you, leaving each one as a draft until you tell it to publish.
Is Nod safe to use?
Nod has real guardrails: a server-side check that asks before uncovered actions, a Vault that keeps passwords out of the model, and a log of every action. But that check is itself an AI model. In n8n's own test of 240 decisions, Claude Haiku 5.5 let 8 unsafe actions through, so pick a strong model, since the check uses it by default, and start with low-stakes tasks.
Is Nod a good OpenClaw alternative?
If you want an agent that runs on your own computer, Nod answers each OpenClaw risk we flagged. It checks steps that could send, submit or change something against your request and rules, types saved passwords through a Vault the model never sees, and logs every action. The checker is still an AI model and can be wrong, and Nod only works while its computer is awake.
How I sourced this: every Nod detail comes from n8n's launch post, Nod's docs and its GitHub README, read on November 11, 2026, when version 0.6.2 was current. I have not run Nod on my own accounts, and I have no affiliate or business relationship with n8n. Full disclosure.