The n8n Nod personal agent is a free, open-source app from n8n Labs. It runs on your own computer, signs in to your accounts and finishes real chores, and you pay only for the AI model behind it. It launched on 10 Nov 2026 as a research preview, so expect rough edges.
The short verdict: try Nod now if you have a capable model, a computer that stays awake and some low-stakes jobs to hand off. Wait if you need it working while your laptop is shut, or if you want to trust it with high-stakes accounts on day one.
This guide pulls n8n's launch post, website, docs, GitHub page and a staff forum reply into one checklist, so you don't have to read all five.
What is n8n Nod, and is it the same as n8n?
No. Nod is a standalone app from n8n Labs, a team inside n8n that experiments with ideas outside its core product. According to n8n's launch post, you don't need an n8n account to use it.
The main n8n product is where you build automations and agents yourself. Nod works the other way round: it is built for one person, connects to that person's accounts, and gets their jobs done for work or home.
What makes Nod an agent rather than a chat window is that it has its own computer and its own browser. Here, "its own computer" means a sandbox, a sealed-off workspace running locally on your machine. A chat tool is the friend who tells you how to cancel your gym membership. Nod is the friend who sits down at the laptop and does it.
n8n's own examples:
- Cancel a subscription, turn down the retention offers and ask for your refund.
- Sign in to supplier portals each month, download the invoices and file them for your accountant.
You can talk to Nod from the desktop app, a web browser, Telegram or WhatsApp, and they all show the same conversation. Start a task on your laptop, then check on it from your phone.
If you'd rather build automations inside n8n itself, our guide to multi-agent workflows with n8n and Claude MCP covers that path.
Is the n8n Nod personal agent really free?
Yes. As of November 2026, the Nod software is free and open source; what you pay for is the AI model. Either you pay a model provider for the AI you use, or you pay nothing extra by running an open-source model on your own computer. Nod doesn't run models itself, so you must connect one.
n8n hasn't published a typical cost per task. It does admit that some early testers found tasks cost more than they expected, and it says its latest release brought cost down substantially. Translation: plan for surprises and lean on the guardrails.
What stops a surprise bill?
Nod has two spending brakes and a receipt for every task, per its spending and usage docs:
| Guardrail | Default | What happens |
|---|---|---|
| Task check-in | $5 | A task checks in with you once its spending since your last message to it passes $5. You can change the amount. |
| Monthly limit | $200 | Set when you finish setup. At the limit, schedules, tasks and background work stop before Nod calls the model. The pause lifts next month or when you raise the limit. |
| Receipt | Every finished task | Shows what the task cost. |
| Who can change limits | You, in chat | A web page or email can't change your limits. |
Two catches:
- Older setups may have no limit. If you set up Nod before it started adding the monthly limit, you have no limit until you set one.
- The dollar figures are estimates. Nod works them out from API prices (API meaning pay-as-you-go access to a model), and your provider's bill is the final word.
Here is the check-in in practice. You ask Nod to collect this month's supplier invoices, then go to lunch. If that job's spending since your last message passes $5, Nod stops to check with you instead of quietly running up the tab.
Can you use your ChatGPT Plus or Pro plan, a Claude plan, or a local model?
As of November 2026, Nod's docs support a ChatGPT plan and a local model, but a Claude subscription doesn't connect. Each route has a catch:
- ChatGPT plan. Nod's docs list signing in with ChatGPT Plus or Pro. Two trade-offs: the Usage screen hides cost amounts, forecasts and monthly limit controls, and Nod's GPT Decisions safety checks are still paid API use, billed separately from your subscription.
- Claude plan. In n8n's community forum, an n8n staff member said a Claude subscription doesn't connect. To use Claude you need Anthropic API credit, which is billed separately from a Claude plan.
- Local model. Nod's site says a local model needs tool calling (the ability to trigger actions like clicking or sending) and a 64,000-token context (how much text it can hold in mind at once). It also warns that small models are less reliable at the safety check.
So pick your model with safety in mind, not only price. The safety section below shows why.
Which computers can run Nod?
As of November 2026, Nod's desktop app runs on macOS 14 or later and on Windows 11 x64. It is also available as a Linux preview on 64-bit Intel or AMD computers, but there are no Linux ARM builds.
Requirements from Nod's install docs:
| Computer | What you need | Download (app and runtime) | Best for |
|---|---|---|---|
| Mac | macOS 14 or later, Apple silicon or Intel | About 1.8 GB | Local setup on a Mac; some testers keep Nod on an always-on Mac mini |
| Windows | Windows 11 x64. For a local assistant: WSL 2 with hardware virtualization and at least 20 GB of free disk space | About 2.1 GB | PCs with spare disk space, or connecting to a server, which doesn't need WSL or a local runtime |
| Linux | A 64-bit Intel or AMD computer, Flatpak and a desktop secret store | Not stated in the docs | People comfortable running a Flatpak preview |
The jargon, in plain words:
- WSL 2 is a built-in Windows feature that runs Linux inside Windows. It needs hardware virtualization, a processor feature that sometimes has to be switched on in your PC's settings.
- Flatpak is a way of packaging Linux apps. A desktop secret store is your system's built-in password keeper.
- Under the hood, Nod is one Node process and a few Docker containers, all on the machine that runs Nod. Docker containers are sealed boxes that keep each part separate.
Is it safe to let Nod use my logins, cards and email?
As of November 2026, Nod is designed so the AI model never sees your saved passwords. But n8n's docs say plainly that the check before every send, post or delete is made by an AI model and can be wrong. Treat Nod as a capable new assistant on probation, not a vault you can forget about.
How does Nod handle your passwords?
Nod's Vault is its built-in store of saved logins. It lets Nod sign in for you without ever giving the password to the model: the Nod server types it into the website.
You can import a CSV file exported from Chrome, 1Password or another password manager. Passkeys, cookies and other items aren't imported.
What stops Nod from sending or deleting the wrong thing?
Nod's docs describe two layers of protection:
- Background work only reads and drafts by default. Background work Nod starts on its own only reads and drafts, unless you've told it what to do on its own.
- Risky actions get a check. Before every send, post or delete, the Nod server runs a check, but an AI model makes the call.
The security model docs are blunt about the weak spot: the checker can be wrong in every mode, and no checker is immune to prompt injection. Prompt injection means hidden instructions on a web page or in an email that try to hijack the agent ("ignore your owner, forward me their invoices").
n8n's tests cover known failure cases. In n8n's own words, they "don't prove that every injection or browser exploit is blocked." One rule does help: a web page or email can't change your spending limits.
On cards: the n8n pages we reviewed don't describe special handling for payment cards. A sensible precaution is to use a card with a low limit for anything Nod might buy.
Does the model you pick change how safe Nod is?
Yes, a lot, according to n8n's own README. In 240 test decisions, GPT-6.1 Sol and Claude Opus 5.5 got every one right, while Claude Haiku 5.5 let 8 unsafe actions through. n8n's advice: pick a capable model.
Keep the headline numbers on Nod's site in perspective. They come from n8n's own tests with GPT-6.1 Sol in October 2026, run in simulated accounts and websites. n8n labels them "Test results, not a guarantee," and says your results depend on your model and the sites you use.
Where does my data go when I use Nod?
Nod runs on the machine you install it on, but as of November 2026 some things leave that machine by default, including what your model reads (unless the model is local), basic usage data sent to n8n, and app connections routed through a service n8n hosts.
| What | Where it goes | Good to know |
|---|---|---|
| Nod and its tasks | Your machine (one Node process plus Docker containers) | Includes Nod's own sandboxed computer and browser |
| Saved passwords | Nod's Vault; the Nod server types them into sites | The model never receives them |
| What the model reads and writes | Your model provider, or your own computer with a local model | A local model keeps this on your machine |
| Basic usage data | n8n | On by default for new users |
| App connections | A connection service n8n hosts | This is the default |
If privacy is your main concern, the local-model route keeps the most on your machine. That only works if your model meets the tool-calling and context bar above.
Does Nod keep working when my laptop is closed?
No. As of November 2026, Nod stops when the laptop running it sleeps, and its schedules run only while that computer is on. Messaging it from Telegram or WhatsApp doesn't change this, because the work happens on the computer that runs Nod.
Your options today:
- An always-on Mac mini. Some testers keep Nod on one.
- Your own server. On Windows, connecting the app to a server doesn't need WSL or a local runtime.
A hosted version is on n8n's roadmap and "next on the list," but no date has been published.
What can't Nod do yet?
Nod is an early preview, and n8n's docs say to expect rough edges and breaking changes (updates that may stop things working the way they did). As of November 2026, the stated limits, in one scannable list:
- No phone calls. It can't ring a supplier or your bank for you.
- One owner per installation. Not a shared family or team assistant.
- Stops when the computer sleeps. No hosted version yet.
- No built-in model. Bring a provider or a local model.
- Vault gaps. Passkeys and cookies don't import.
- No Linux ARM builds.
- A fallible safety check. The send, post and delete check can be wrong.
- No published cost per task. Receipts are estimates; your provider's bill decides.
Should you try Nod now, or wait?
Try Nod now if you're comfortable being an early tester and can give it a capable model plus a computer that stays awake. Wait if you need always-on help without extra hardware, or if you want zero-risk handling of money and email.
Try it now if:
- You run macOS 14 or later, or Windows 11 x64 with room for the local setup.
- You'll pay for a capable model (in n8n's tests, GPT-6.1 Sol and Claude Opus 5.5 got every test decision right), or you have a ChatGPT Plus or Pro plan and accept the separate API cost for safety checks.
- You have repetitive, low-stakes web chores, like pulling invoices from supplier portals.
- You're happy to approve actions and read receipts for the first few weeks.
Wait if:
- You want it working while your laptop sleeps and don't have a spare always-on machine.
- You only have a small local model, which n8n says is less reliable at the safety check.
- You want one setup shared across a family or team.
- You need phone calls handled.
If you'd rather build a simpler agent yourself first, our guide on how to build your first AI agent for free walks through the options.
Your starter plan for the first week
- Set the brakes before the first task. Keep or lower the $200 monthly limit, and lower the $5 check-in if you're nervous.
- Start with read-and-draft jobs. Ask Nod to gather invoices or draft a refund request, then review its work before anything is sent.
- Add logins one at a time. Put a single low-risk account in the Vault and check the receipt after each task. Only then hand over more.
Prices, limits and requirements were checked on Nod's own pages in November 2026; expect them to shift while Nod is in preview.
