Nod by n8n is an open-source personal AI agent that runs on your computer and uses a model you choose. It can work across your apps and websites, remember useful context and run recurring tasks. But does it offer anything you cannot already do with Claude Code or Codex? That depends less on the feature list and more on how much of your own AI setup you want to build.
Last updated: 2026-11-17
What Is Nod, And Who Is It For?
Nod is a standalone personal agent from n8n Labs, the experimental team inside n8n. It is available as a research preview, and you do not need an n8n account or an existing workflow to use it.[1]
Instead of building an automation step by step, you describe the result you want. Nod then works out how to tackle it using connected apps, its own browser and a separate computer environment.
Think preparing a client-call brief, chasing an invoice or comparing venues for an offsite. Those are examples n8n promotes, not proof that every task will work reliably on every website.[1]
The three things that make Nod interesting are:
- Open-source code: Nod uses the Apache 2.0 licence.
- Model choice: You can connect one of more than 30 supported providers, a supported ChatGPT plan or a suitable local model.
- Your own host: The runtime lives on your computer or a server you control.[1]
That last point is both a benefit and a responsibility. You choose where Nod runs, but you also need to keep that machine available.
Here is the short version:
| Question | Answer |
|---|---|
| What is it? | A standalone, open-source personal AI agent from n8n Labs, currently a research preview. |
| Where does it run? | On your computer or a self-hosted server. The host must stay on and awake. |
| Which computers are supported? | macOS 14 or later; Windows 11 and Linux x64 Flatpak are available as previews. |
| Which AI can it use? | Supported model providers, ChatGPT Plus or Pro sign-in, or a suitable local model. |
| What does it cost? | The app is free. Model-provider fees or supported plan limits still apply. |
| Do you need n8n? | No. Connecting an existing n8n instance is optional. |
| Does local mean offline? | No. Models, apps and websites receive the data needed for the task. |
| What is not available yet? | Managed hosting, phone calls and shared team use are on the roadmap. |
These availability and roadmap details come from Nod’s current product FAQ.[1]
What Can Nod Actually Do?
Nod combines app connections, a browser, memory and recurring work in one interface. The point is not just to answer questions, but to carry a task across the tools needed to finish it.[1]
Work Across Apps And Websites
You can connect tools such as Gmail, Slack, Notion, Drive and Calendar. Nod also has its own browser for work on websites, rather than relying only on an app’s available actions.[1]
A useful starting point is a defined outcome: prepare a brief for tomorrow’s client call using the calendar entry, relevant emails and project notes.
That is more helpful than “organise my work”. It gives the agent a clear target and gives you something concrete to check.
If Notion already holds your projects, tasks, documents and working instructions, you have a head start. The agent has a coherent place to find context instead of piecing together five different versions of what matters.
Remember Context And Repeat Work
Nod has a memory system you can inspect, edit and remove entries from. It can remember preferences and learn procedures for tasks, rather than starting from scratch each time.[1]
That is one of the more interesting promises here. Getting a model to do a task once is useful; avoiding the same explanation every week is where a personal agent starts to earn its place.
You can also schedule recurring work, such as a regular brief or a check for changes on a website. Each run can use model resources, so a frequent schedule is not automatically a free one.[1]
I would not judge the memory system from a launch tour, though. The real test is whether it remembers the right things and stays useful after weeks of actual work.
Chat From Telegram Or WhatsApp
You can use Nod’s app directly or connect Telegram or WhatsApp to the same conversation, memory and tasks. The machine running Nod still needs to be awake and connected.[2]
The WhatsApp QR connection uses an unofficial client. Nod’s docs warn that it can break after WhatsApp updates and carries a risk of account restrictions.[2]
Convenient? Yes. Something I would connect without reading the caveats? No.
How Do You Get Started, And What Does It Cost?
You get started by installing Nod, connecting a model and giving it a small, clearly defined job. The app is free, but the AI usage comes from your chosen provider or supported subscription.[1]
The basic setup is:
- Download the appropriate app from the Nod website.
- Choose your model connection during onboarding.
- Connect only the apps needed for your first task.
- Ask for a specific outcome.
- Check the result and review Usage before expanding the setup.[1]
For ChatGPT Plus or Pro users, the supported Sign in with ChatGPT option may be the simplest starting point. It uses your plan’s included usage; it does not give you unlimited extra capacity.[3]
If you want Claude, you need an Anthropic API connection rather than your Claude subscription login. Other supported providers have their own connection methods and billing terms.[3]
Local models are another option. Nod specifies tool calling and a 64,000-token context, and warns that smaller models are less reliable at its safety check. “Supports local models” is not the same as “any small model will work well”.[1]
Pro Tip: Start with a public-web research task before connecting your whole working life. You can check how Nod handles instructions without immediately granting access to sensitive accounts.
What Happens When Your Laptop Sleeps?
Nod cannot keep working on a sleeping host. For background tasks while your main laptop is closed, run it on an always-on computer or a self-hosted server.[1]
My plan is to try it on my Mac mini. That should let me test whether the recurring work sticks, rather than making availability depend on my laptop being open.
Managed hosting is on the roadmap, not something to assume is included today.[1]
Does Running Nod Locally Keep Your Data Private?
Running Nod locally keeps its database, files, memory and saved logins on your chosen host. It does not mean everything stays on that machine or that your tasks run offline.[1]
The important distinction is between where the agent runs and where task data goes:
- A hosted model can receive your conversation, information Nod reads and screenshots.
- Connected apps receive the actions Nod sends to them.
- Websites receive the information submitted to them.
- Default app connections use n8n’s connection service and Composio.[4]
Usage telemetry starts on, while conversation traces start off. You can control both in the privacy settings; usage telemetry excludes your messages and Nod’s replies.[4]
This is why “runs locally” should not become shorthand for “all data stays in Europe” or “nothing leaves my computer”. Your model, connections and destinations still matter.
Nod is available in the EU, UK and Switzerland, subject to your AI provider’s terms. That is useful access information, not a blanket data-residency guarantee.[1]
How Do Approvals And Vault Work?
Nod checks proposed actions against your request and rules, and asks when those instructions do not cover the action. Vault separately handles supported secrets so the model does not need to see them.[5][6]
That does not mean you approve every click.
If you ask Nod to move a particular meeting and tell the attendee, that request can authorise both the calendar change and the message. A request to search for flights does not, by itself, authorise buying one.[5]
You can add rules such as asking before sending external emails. But the permission checker still makes a model-based judgement, and n8n explicitly says it can be wrong.[5]
There is a useful boundary here, not a guarantee that an AI agent cannot make mistakes.
What Happens When Nod Pays For Something?
For a card saved in Vault, Nod requires approval of that specific payment and asks you to enter the card’s security code in its private payment form. It never saves the code or exposes the card details to the model.[6]
Telegram and WhatsApp cannot approve the payment or receive that code. You need to open Nod’s web or desktop interface.[6]
That adds friction, deliberately. You cannot treat a saved card as permission for unattended purchases.
Can Nod Build n8n Workflows?
Yes. Connect an existing n8n instance and explicitly ask Nod to build an n8n workflow. It can create the workflow, test it with sample data and show you the result; a newly built workflow stays a draft until you ask it to publish.[7]
For example, you could ask for an n8n workflow that checks a condition every weekday and posts a Slack update when it is met.
Sample testing is not the same as a successful live run. Nod reports what it tested and what still needs setup, such as credentials.[7]
My scepticism here is fairly simple: if AI can help me write actual code, being able to generate a visual workflow is not enough on its own to make me switch back to no-code.
But if you already run n8n and value its visual builder, this connection makes sense. You can delegate the build while keeping the result in the system you already use.
Should You Try Nod?
Nod is worth trying if you want a personal agent without assembling the browser, memory and scheduling pieces yourself. It is less obviously compelling if you already have a setup that does those jobs well.
I would put you in the “give it a go” group if:
- You want to delegate tasks, not just chat about them.
- You value open-source software and model choice.
- You are comfortable keeping a host running.
- You are willing to test an early product with bounded tasks.
If you already have Claude Code or Codex working across your files and tools, the extra value may be smaller. The question is whether Nod removes enough setup and maintenance to justify another interface.
And if you already use a personal agent you like, I would not switch just because another one launched.
For me, the interesting test is not whether Nod can cancel a subscription once. It is whether I keep handing it useful work, or go back to my existing Codex-led setup.
The open-source approach and model choice make it worth exploring. The research-preview label means you should explore with your eyes open.[1]
Frequently Asked Questions
Is Nod Free To Use?
Nod is free to download and its code uses the Apache 2.0 licence. Model usage is separate: provider fees apply, or you use the included allowance of a supported plan such as ChatGPT Plus or Pro.[1]
Do You Need An n8n Account To Use Nod?
No. Nod is a standalone app, not the n8n AI Assistant, and does not require an n8n account. Connecting your existing n8n instance is optional if you want Nod to build workflows there.[1][7]
Can Nod Use Claude Or A Local Model?
Yes. Claude uses an Anthropic API connection, while suitable local models can connect through an OpenAI-compatible server. Local models need tool calling and a 64,000-token context, and smaller models can make less reliable safety decisions.[3][1]
Does Nod Work When Your Computer Is Off?
Not if that computer hosts your Nod runtime. For work while your main computer is off or asleep, run Nod on a separate always-on machine or server. Managed hosting is currently on the roadmap.[1]
Does Nod Ask Before Every Action?
No. Your request and standing rules can authorise actions without another prompt, while actions outside that scope require confirmation. A saved-card payment has its own explicit approval and security-code step; the general permission checker is still model-based and can make mistakes.[5][6]


