n8n Nod: The Free, Open-Source Personal Agent That Runs on Your Own Computer
A walkthrough of Nod, n8n Labs' answer to Duet, DOTS, Muse and Grokbot, covering the local-first design, the four safety guardrails, setup, and the weak spots.
Posted
1 weeks ago
Duration
Format
Review
educational
Views
33K
364 likes
57 · 43
Big Idea
The argument in one line.
Nod is the first serious personal agent that runs on your own machine with any model you choose, trading the always-on convenience of cloud agents for privacy, cost control and open code.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
A small business owner who wants an agent to triage email, send weekly recaps and watch competitor pages, but doesn't want files and logins living on someone else's server.
Someone already paying for a ChatGPT plan who wants an agent that uses that monthly usage instead of billing per task.
An n8n user deciding how a judgment-based agent fits next to fixed, repeatable workflows.
A technical user who wants to read, fork or self-host an agent, or point it at a local Ollama or LM Studio model.
SKIP IF…
You want an agent you can message from your phone with zero setup and have it keep working while your laptop is closed. The cloud agents fit that better.
You need hard payment limits at the bank level, such as single-use card numbers tied to one store and one amount.
TL;DR
The full version, fast.
n8n Labs released Nod, a free, Apache 2.0 personal agent that creates a private Linux computer inside your own machine and does tasks there. You pick the brain: a ChatGPT plan sign-in, an API key from OpenAI, Anthropic, Google or OpenRouter, or a local model through Ollama or LM Studio. Four guardrails apply: a Vault so the model never sees passwords, server-side approvals, a $200 default monthly cap with a $5 per-task pause, and an Activity Log. Setup is a roughly 2 GB download. The tradeoffs are that it sleeps when your computer sleeps, costs can climb, payments are less locked down than some rivals, and it is a single-owner research preview.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Duet, DOTS, Muse and Grokbot all do the same job. n8n's Nod is free, open source, and runs on your computer with any model.
00:49 – 01:41
02 · What Nod is
A personal agent from n8n Labs, separate from n8n itself, Apache 2.0, free apart from the model, and still a research preview.
01:41 – 04:25
03 · Local computer vs cloud agents
Where the agent's computer lives, what happens when your laptop sleeps, and the three ways to choose a model.
04:02 – 05:25
04 · The four guardrails
Vault, server-side approvals, spending limits and the Activity Log.
05:25 – 07:21
05 · Install and setup
Download from nod.n8n.io, onboarding questions, model sign-in, and the long private-computer setup.
07:21 – 08:41
06 · A tour of the app
Main chat, Schedules, Apps, Channels for Telegram and WhatsApp, Usage and Vault.
08:41 – 09:37
07 · What it's good for
Inbox triage, weekly recaps, competitor watchers, vault-backed logins, and splitting work with n8n.
09:37 – 11:44
08 · Weak spots and who it's for
Sleeps with your computer, costs, payment limits, single owner, early preview, then a verdict and a guide pitch.
Atomic Insights
Lines worth screenshotting.
The real difference between Nod and cloud agents is where the agent's computer lives: on your machine instead of on a vendor's server.
Nod runs a small private Linux computer inside your own machine, so the agent works in a sealed space and can't wander through the rest of your files.
If your laptop sleeps, a local agent sleeps too, even when you message it from Telegram. Cloud agents keep working while your laptop is closed in your bag.
Signing in with a ChatGPT plan you already pay for means Nod draws on your plan's monthly usage and adds no per-task bill.
A local model only works with Nod if it supports tool calling and has at least 64,000 tokens of memory, which rules out very small models.
In Nod's Vault, the AI model never sees your password. It only knows a login exists, and Nod's server types the password into the site.
Nod's approval check runs on its server, outside the model, so the rule still holds if the model gets confused or a webpage tries to trick it.
Nod defaults to a $200 monthly spending limit and pauses any single task that spends more than $5 to ask whether to continue.
Agents that browse websites burn a lot of tokens, which is why some early testers spent more than they expected.
Nod can't cap the amount at your bank, and while it types a saved password, Nod's own server can read it even though the model can't.
Pair Nod with n8n by giving the agent the messy one-off jobs that need judgment and leaving the fixed workflows to n8n.
A good first Nod task is something boring you do every week, such as a Friday recap across email, calendar and Slack.
Takeaway
Local agents trade convenience for control
WHAT TO LEARN
A personal agent that runs on your machine gives you privacy, model choice and cost visibility, and you pay for it with setup time and an agent that sleeps when your computer does.
02What Nod is
Nod is a separate app from n8n Labs under the Apache 2.0 license. It needs no n8n account and no workflow knowledge, and it's still a research preview.
03Local computer vs cloud agents
Cloud agents run on a vendor's server and models. Nod runs a private Linux computer on your machine, so your files and browser sessions never leave it.
The cost of that privacy is uptime: if your computer sleeps, Nod stops, unless you give it an always-on Mac Mini or your own server.
You can power Nod with a ChatGPT plan, an API key from OpenAI, Anthropic, Google or OpenRouter, or a local model that supports tool calling and 64,000 tokens of memory.
04The four guardrails
Keeping credentials away from the model is a design pattern: the Vault stores logins and Nod's server types them in, so a confused or tricked model can't leak a password.
Safety rules enforced outside the model, such as approvals before sending, paying, posting or deleting, still hold when the model misbehaves or a webpage injects instructions.
Set a low monthly spending limit on day one. Nod defaults to $200 and pauses any task over $5, because browsing agents burn tokens quickly.
05Install and setup
Setup downloads about 2 GB for the private computer. n8n quotes 10 minutes but it took close to half an hour on a MacBook, so start on a good connection.
06A tour of the app
The main chat plans the work and hands it to tools, with tabs for Schedules, Apps, Channels, Usage and Vault. Telegram and WhatsApp let you reach it while you're out.
07What it's good for
Good fits are recurring, low-stakes jobs: a daily inbox triage with drafted replies, a Friday recap, a competitor price watcher, and invoice downloads from a supplier portal.
Split the work with n8n: let the agent handle messy one-off jobs that need judgment and let workflows run the fixed processes that must repeat identically.
08Weak spots and who it's for
Weigh the weak spots before committing: it sleeps with your computer, payments lack bank-level limits, it serves one owner, and it is still a research preview.
Glossary
Terms worth knowing.
Personal agent
An AI that takes a task, then opens apps and websites on its own computer to get it done, instead of only answering in a chat.
Nod
A free, open-source personal agent from n8n Labs that runs a private Linux computer on your own machine.
n8n Labs
The experimental team at n8n, the automation platform. Nod is a separate app that needs no n8n account.
Vault
Nod's store for logins and cards. The model only learns that a login exists, and Nod's server enters the password itself.
Tool calling
A model's ability to trigger outside actions such as using an app or browser. A local model needs it to work with Nod.
OpenRouter
A single API key that gives access to hundreds of AI models, billed by usage.
Ollama / LM Studio
Programs that run AI models locally on your own computer so nothing leaves the machine.
MCP server
A connector that lets an AI agent use an outside tool or service. Nod lets technical users add their own.
“When you install it, the loading screen literally says, starting your private computer.”
Vivid, self-contained explanation of the local-first idea→ IG reel cold open↗ Tweet quote
04:18
“When Nod needs to sign into a website, the AI model never sees the password.”
One-line safety claim people want to repeat→ TikTok hook↗ Tweet quote
11:00
“Nod handles the messy one-off jobs that need judgment, and n8n runs the fixed workflows that must happen the same way every time.”
Clean mental model for agents plus workflows→ newsletter pull-quote↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
In the last few weeks, almost every big AI lab has released its own personal agent. Anthropic has CloudDuet, OpenAI has DOTS, Meta has Muse, and XAI has Grokbot. They all do roughly the same job.
You give them a task, and they go off on their own computer, open your apps and websites, and get it done. Now, N8n, the automation platform, has released one too. It's called Nod.
and at first it looks like a late copy of something we already have five times over. Two things set it apart. Nod is free and open source, and it runs on your own computer with whatever AI model you want, including one that never leaves your machine.
In this video, we'll cover what Nod is, how it differs from Duet and DOTS, how it handles your passwords and your money, and how to install it. Then we'll walk through the app, see what it's good for, and finish with its weak spots, some of which NA10 admits itself. Nod is a personal agent made by NA10 Labs, the experimental team at NA10.
It's a separate app from NA10 itself. You don't need an NA10 account, and you don't need to know anything about workflows to use it. It's still a research preview, so expect some rough edges.
The code is public on GitHub under the Apache 2 .0 license, so you can read it, change it, and build your own version on top of it. The app costs nothing. You only pay for the AI model behind it, and even that can be free.
The biggest difference with the other agents is where the agent's computer lives. Take DOTS as an example, though Duet, Muse, and Grokbot all work the same way. You send a message from the ChatGPT app.
That message goes to OpenAI's servers. Over there, DOTS has its own virtual computer in the cloud, with its own browser, and it runs on OpenAI's models.
It does the work in your apps and on websites, and sends you the result. Everything happens on someone else's server. NOD follows the same steps, except the computer is yours.
You send a message from the NOD app on your desktop, or from Telegram or WhatsApp on your phone. That message doesn't go to N8n's servers. Nod sets up a small private Linux computer right inside your own machine, with its own files and its own browser, and the work happens there.
When you install it, the loading screen literally says, starting your private computer. The upside is privacy. Your workspace, your browser sessions, and your files stay on your machine, and the agent works inside its own sealed -off space, so it can't go wandering through the rest of your computer.
The downside is that if your computer is off, Nod is off. If your laptop goes to sleep, Nod stops, even if you message it from Telegram.
The cloud agents keep working while your laptop is closed in your bag. Nod doesn't, unless you give it a machine that never sleeps. People do that with a Mac Mini that stays on all the time, or by installing Nod on their own server, which N8n supports and documents.
There's also no Nod app on your phone for now. To reach it while you're out, you connect a Telegram bot or your WhatsApp, and you talk to your computer through that.
The other big difference is the brain. DOTS runs on OpenAI's models, Duet on Anthropix, and Muse on Metas. Nod lets you choose from three kinds of model.
Option one is to sign in with a chat GPT plan you already pay for. Nod then uses your plan's monthly usage, so you don't pay anything extra per task. That's what we picked during setup, and it connected to GPT -6 Sol in a few clicks.
Option 2 is an API key from OpenAI, Anthropic, Google, or OpenRouter. OpenRouter alone gives you access to hundreds of models, and you pay for what you use. Option 3 is a model that runs locally on your own computer, through tools like Ollama or LM Studio.
In that case, nothing leaves your machine, and the only cost is electricity. The model has to support tool calling and needs a memory of at least 64 ,000 tokens, so very small models won't work. and you'll want a fairly powerful computer.
Put together, Nod does about 90 % of what Duet and DOTS do, on your own machine, with any model, and with code anyone can read and change. An agent that logs into websites and can use your card needs guardrails, and Nod has four. The first is the Vault.
This is where you save your logins and your cards. When Nod needs to sign into a website, the AI model never sees the password. it only knows that a login exists for that site.
Nod's server types the password into the website itself, so it's never sent to the model provider. For payments, you approve every purchase yourself with your card's security code, and Nod never saves that code. The second guardrail is approvals.
Before Nod sends, pays for, posts, or deletes anything you didn't ask for, it stops and asks you first. This check runs on Nod's server, outside the AI model. so the rule still applies if the model gets confused or a webpage tries to trick it.
The third is spending limits. When you finish setup, Nod puts a monthly limit of $200 on your AI usage. You can change it to any amount from $1 up.
When you reach it, every task stops until the next month. Separately, if a single task spends more than $5, Nod pauses, tells you the cost, and asks if you want to keep going. The fourth is the Activity Log.
For every request, it lists each step the model took, with its tokens and cost, and why each action was allowed. You can export it to a spreadsheet. To install it, go to nod .n8n .io and download the app.
On a Mac, you need macOS 14 or newer, and it works on both Apple chips and Intel. Windows is supported too. Open the file, drag Nod into your Applications folder, and launch it.
The welcome screen sums up what it does. It gets tasks done, works proactively, runs on your computer, and keeps you in control. Click Get Started, and Nod starts setting up your private computer in the background.
While that's running, it asks you a few questions so it can be useful from day one. Do you want help with work, your personal life, or both? What do you do?
Is it for you or for your clients? How big is your company? And what would you like to take off your plate?
You can pick things like staying on top of email and Slack, handling recurring tasks, or managing your calendar. Next, you choose your model. Here, we signed in with a chat GPT plan.
A sign -in page opens, you approve it, and you're connected. Then, Nod asks if you want to share your conversations with the NA10 team to help improve it. It's your choice, and you can turn it off later in the settings.
Then comes the longest step. Starting your private computer. NA10 says the whole setup takes about 10 minutes.
On our MacBook, it took close to half an hour because it downloads around 2 gigabytes the first time. Start it on a good connection and go grab a coffee. You only do this once.
When it's done, you land on the main screen, which looks like any AI chat. You talk to this main chat and it plans the work and hands it off to the right tools. Nod already suggests a few tasks based on your answers, like...
Brief me on my day every morning, send me a recap every Friday, or scan the news in my field every Monday. Some of them have a small label that says repeats. Those are recurring jobs.
On the left, you have the rest of the app. The first tab is schedules. These are tasks that run by themselves at a set time.
You can create one by clicking a suggestion or just by describing it. Nod asks you one question at a time, what to do and when. then shows you exactly what will run before you approve it.
From there, you can open its details, edit it, ask Nod about it, or run it right away. If you've used scheduled tasks in Claude, this will feel very familiar. The next tab is Apps.
These are the tools Nod can use for you, like Gmail, Google Calendar, Drive, Slack, Notion, and Linear, plus a long list of others. Most of them connect in one click, and if you're more technical, You can also add MCP servers and your own APIs.
Then there's channels. This is where you connect Telegram or WhatsApp. For Telegram, you create a bot with Botfather and paste in its key.
For WhatsApp, you scan a QR code with your phone. After that, it's the same conversation, whether you're at your desk or out of the house. Usage holds the cost log and the spending limit.
The last tab is the Vault, with your saved logins, your cards, and the websites where Nod is still signed in. Nod is good at the same tasks you'd give Duet or Dots, and it does them about as well. Here are some ideas that work for a small business.
Every morning, have Nod go through your inbox, flag what needs a reply, and draft those replies for you to check. Every Friday, have it send you a recap of what happened that week across your email, calendar, and Slack. Use it as a watcher.
Ask it to check a competitor's pricing page or a supplier's product page every day and message you on Telegram the moment something changes. Hand it the boring logins. If you download invoices from a supplier portal every month, save the login in the vault, and Nod can sign in, download the files, and put them in your drive, without the model ever seeing your password.
If you already use N8n, you can connect Nod to your N8n account and split the work. Nod handles the messy one -off jobs that need judgment, and N8n runs the fixed workflows that must happen the same way every time. Now, the weak spots.
Some of these come straight from N8n's own documentation, which is honest of them. 1. If your computer sleeps, Nod sleeps.
To run it 24 -7, you'll need a machine that stays on, or your own server. 2. Costs can add up.
N8N says some early testers spent more than they expected because agents use a lot of tokens when they browse websites. Set a low monthly limit on day one. And if you have a chat GPT plan, signing in with it is the easiest way to keep costs under control.
3. Payments are less locked down than on some competitors. Meta's Muse uses single -use card numbers tied to one store and one amount.
Nod can't limit the amount at your bank, and while it's typing in a saved password, Nod's own server can read it, even though the AI model can't. It's built for one person.
Each install has one owner, and it can't make phone calls. It's an early preview. There's no bug bounty program yet, and you'll run into rough edges.
So who is Nod for? If you want an agent that just works from your phone with zero setup, the cloud agents like Duet and DOTS are still the easier choice. But if you want to pick your own model, keep your files and passwords on your own machine, see exactly what every task costs, or even change the code yourself, Nod is the first serious agent that gives you all of that.
And it's free. If you try it, set a spending limit first, then give it one boring task you do every week and see how far it gets. Before you go, we put everything from this video into one complete guide.
It walks you through the install step by step, the settings to change in your first 10 minutes, how to pick the right model, and how to connect Telegram and WhatsApp. You'll also find ready to paste prompts for your daily life, your studies, and your business, and what to do when something goes wrong.
The link is in the description.
The Hook
The bait, then the rug-pull.
Every major AI lab now ships a personal agent that runs your apps for you, and n8n just added one that looks like a late copy. Two things set Nod apart: it's free and open source, and the agent's computer is yours.
Frameworks
Named ideas worth stealing.
04:02list
Nod's four guardrails
Vault
Approvals
Spending limits
Activity Log
Credentials hidden from the model, server-side approval before sending, paying, posting or deleting, a monthly cap plus a per-task pause, and a per-step cost log.
Steal fordesigning safety for any agent that touches logins or money
03:04list
Three ways to power Nod
Sign in with a ChatGPT plan
API key (OpenAI, Anthropic, Google, OpenRouter)
Local model (Ollama, LM Studio)
Plan usage means no per-task cost, API keys are pay-as-you-go, and local models keep everything on the machine.
Steal forchoosing a model for any BYOK agent
CTA Breakdown
How they asked for the click.
VERBAL ASK
11:29product
“We put everything from this video into one complete guide. The link is in the description.”
A soft pitch at the end, after a clear verdict and a concrete first step, so the guide reads as a continuation of the video.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
A 14-minute walkthrough of the Noose desktop installer that finally lets non-technical users run one of the most capable open-source AI agents without touching a terminal.
How Jack Roberts wired Google NotebookLM into the Hermes Telegram agent -- turning a free 300-source research platform into a 24/7 AI assistant that reads, acts, and automates from your phone.
A ten-minute rapid-fire tour of thirteen real things people are doing with Claude, Grok bots and Google AI Mode right now, from haggling with Comcast to coaching your tennis swing.