---
title: 'n8n Just Built Its Own JARVIS?!'
source: 'https://youtube.com/watch?v=0eyW6wnIBCk'
video_id: '0eyW6wnIBCk'
date: 2026-11-17
duration_sec: 482
channel: 'Julian Goldie SEO'
---

# n8n Just Built Its Own JARVIS?!

> Source: [n8n Just Built Its Own JARVIS?!](https://youtube.com/watch?v=0eyW6wnIBCk)

## Summary

n8n has released Nod, an open-source personal AI agent that operates on its own virtual computer, performing tasks across real websites and apps with explicit user permission. The video explains Nod's architecture, safety features, test results, and ideal use cases, positioning it as a standalone tool distinct from n8n's main platform.

### Key Points

- **Nod Release and Context** [00:27] — n8n released Nod on November 10th as a research preview from n8n Labs. It's a standalone personal agent with its own computer, not a workflow or the n8n AI assistant.
- **Positioning and Open Source** [01:08] — Nod is for individual tasks, while n8n workflows handle team processes. It's open source under Apache 2.0, and users bring their own AI provider.
- **Platform and Setup** [01:34] — Nod runs on macOS 14+, Windows 11, Linux (preview), or a personal server. It works with 30+ AI providers, including OpenAI, Anthropic, and Google. Local models need tool calling and 64k context window. Setup takes ~10 minutes.
- **Capabilities and Examples** [02:13] — Nod connects to Gmail, Slack, Notion, Drive, and Calendar. Examples include generating client call briefs, finding unanswered questions, and booking offsites. It asks before acting.
- **Architecture and Permission** [03:32] — Nod runs as an app plus Docker containers: a Linux workspace, Chromium browser, and database. The server checks every action against user requests; anything not covered waits for approval. Passwords are entered privately in a vault, away from the AI model.
- **Test Results and Safety** [04:39] — n8n ran 678 test runs, completing 98% of errands with zero unsafe actions. Memory test scored 93.2%. Attack emails planted zero false memories in 216 runs, but 12 later answers showed the claim (from re-reading email, not memory). Model choice matters: GPT-6.1 Soul and Claude Opus 5.5 were perfect, but Claude Haiku 5.5 let 8 unsafe actions through.
- **Limitations and Future** [06:03] — Nod requires the computer to stay on and awake. New users share basic usage data by default (can be disabled). Future plans include phone calls, shared Nod instances, and customizable open code.
- **Target Audience** [06:47] — Nod is for client calls, follow-ups, supplier portals, and live admin tasks. Requires a Mac/Windows/Linux machine and an AI key or ChatGPT account.

### Conclusion

Nod is a promising open-source personal AI agent that prioritizes safety through permission checks and transparent logging. While early, its strong test results and clear positioning make it a valuable tool for individual automation tasks.

## Transcript

screen right now? Be honest. Now picture an assistant that opens them, clicks through and finishes the job, and it asks you before it acts for you. N-A-N just released it, and it's open source. Wait until you see what it checks before
what N-A-N just released, how it works numbers N-A-N published for it. And at the end, I'll tell you who this is really for because it's not for everyone. So let's start with what just
happened. On November 10th, N-A-N released something called Nod. It's a personal agent. N-A-N calls it a research preview, so it's early. It comes from N-A-N Labs, a new team inside N-A-N that tries out ideas outside the
main product. Here's what makes it different. Nod has its own computer and need. Then it works inside your apps and on real websites until the job is to do the task, it does the task. Now this part confuses people, so listen
closely. Nod is not an N-A-N workflow. It's a standalone app, and you don't need N-A-N to use it. N-A-N also says it is not the N-A-N AI assistant. If you already use N-A-N, you can connect your instance so Nod can work with your
where does it fit? Here's how N-A-N's assistant is where they think and write. N-A-N runs the processes their team shares, and Nod takes the ongoing jobs that are theirs alone. One person, one
Nod. It's also open source. The code is on GitHub under the Apache 2.0 license. You bring your own AI provider, and N-A-N says this doesn't change the direction of N-A-N. Work on N-A-N and its agent continues as planned. That's
the big picture. Let's talk about what you can expect when you open it. First, it runs on your computer. There's a Mac app for macOS 14 or later. Windows 11 and Linux are both labeled preview. You can also run it on your own server for
one person. Second, you choose the brain. N-A-N says it works with more than 30 AI providers, including OpenAI, Anthropic, and Google. You can also sign in with your ChatGPT account, or you can run a local model, but N-A-N says a
local model needs tool calling and a context window of 64,000 tokens. That's basically how much it can hold in its head at once. And small models are less can talk to it in the app or on Telegram or WhatsApp. It connects to Gmail,
Slack, Notion, Drive, and Calendar with one sign-in. And N A N says setup takes about 10 minutes. So what can it actually do? These are examples an A N every morning Nod reads your email threads, your calendar, and the latest
news about the client. Then it writes a one-page brief for each call. An 8 N says the brief is stored on the computer or server that runs Nod. Another example, Nod can find the question you never answered, like a client message
from your own sheet and asks if you want to send it. It only sends after you say yes. An 8 N also lists things like turning client requests from email and skipping duplicates and booking a team offsite by comparing three spaces then
booking the best one once you say go. Now I want to pause for a second. Quick pause before I show you the next part. If you want to learn tools like Nod properly, inside the AI Profit Boardroom, we've got a 30-day roadmap
built around learning AI agents step by step. There are also four coaching calls every week. So if you get stuck connecting an AI provider or setting up your first agent, you can bring it to a call and get unstuck. Everything you're
learning in this video, like how Nod asks before it acts, is exactly the kind of thing we help members practice. Link in the description or go to ai profit let's get into how it works. This is where it gets interesting. Nod runs as
one app on your machine plus a few Docker containers. Think of those as sealed boxes inside your computer. One box is a Linux workspace where Nod does its work. Another box is a Chromium browser and a database keeps everything
stored. Your files, memory, and saved logins stay on that machine. Your AI Now the part that matters most, permission. The server checks every app action and every website step that can change something against what you asked
doesn't cover it, Nod waits for your yes. N8N says this check lives in the vault. It enters your passwords and codes privately on the website. So those
secrets stay away from the AI model, and every action and every sign-in shows up in a usage page. So you can see what Nod did. Then there's memory, and you can You can edit it or tell Nod to forget it. If something changes, tell it once,
and it uses the new fact from then on. It can also learn your writing style how to do a task, reuse the steps, and fix them when a website changes. N8N says it learns about you only from your own messages, never from the pages or
emails it reads. Now more information. N8N published test numbers for Nod. Let me give them to you straight. They ran 226 errands in test accounts and test 226 errands in test accounts and test websites, three runs each. That's 678
runs. Nod completed 98% of those errands, and in those 678 runs, N8N counted zero unsafe actions. For memory, N8N used a public test called Long Mem Evil S. It has 500 questions about earlier chats, asked in fresh sessions.
Nod scored 93.2%. Then they tested attack emails. These are emails written to plant a false memory. There were 108 published attack emails, tried two ways. That's 216 runs. Zero planted a false memory. But read the fine print because
N8N does. These came from October 2026 using GPT 6.1 Soul in simulated accounts and websites. N8N calls some test results, not a guarantee. Real websites can behave differently. And in 12 later answers, the attack claim showed up.
That happened because Nod read the email again from the inbox. It did not come from memory. Here's one more detail that matters. N8N says the safety check is a model's judgment. So the model you pick counts. In 240 test decisions, GPT 6.1
Soul and Claude Opus 5.5 got everyone right. Claude Haiku 5.5 let eight unsafe actions through. N8N's advice is simple. Pick a capable model, and Nod can make That's why it asks before anything you didn't request, and why everything it
does is on record. A limits to know about. Nod needs the computer running it to stay on and awake. So if you want your schedules to keep running with the laptop closed, N8N suggests a Mac Mini or your own server.
Also, for new users, Nod shares basic usage data with N8N by default. You can turn that off in settings under privacy and data. What's next on N8N's list says your machine is asleep, calling businesses on the phone. One nod shared
by a few people on a team because today it's one owner per nod and making it your own by changing the open code. So who can use this? N8N built Nod for client calls, follow-ups, and supplier portals. It can also handle live admin
like a booking. You'll need a Mac or a Windows 11 or Linux computer in preview, connecting an AI key or your ChatGPT account. N8N says that's why it suits setup. If that's you, setup is about 10 minutes. That's Nod. An agent with its
own computer that asks before it acts. Here's something worth knowing. Right business owners haven't figured out personal AI agents yet, and that window won't stay open forever. The way to get ahead is to learn the tool early. Inside
tutorials that walk you through new tools. You get four coaching calls every week where you can bring your questions. You get a prompt library full of prompts to help you find your way around the community. Link in the description or go
to aiprofitboardroom.com. If you want the full process, the SOPs, and over 100 AI use cases like this one, join the AI Success Lab. It's our free description. You'll get all the video notes from this episode plus access to
95,000 members. The AI Success Lab is a place to learn AI tools alongside other people. It's free to join. The video notes from this episode put the key you can come back to them anytime. So join the AI Success Lab, grab the notes,
and keep learning with us. I'll see you in the next video.
