n8n Releases Nod, an Open-Source Agent That Runs on Your Computer

Most people who have tried to automate routine work run into the same wall. If there is an API, everything works. If there is no API, just a personal account area with three forms and a captcha, you have to do it by hand. That is exactly the gap Nod targets, a new open-source agent from n8n Labs. The announcement drew more than a million views on X, and it is not only because of the big name.

What was shown

The post reveals little, but the main point is stated plainly. Nod is a personal open-source agent. It runs on your computer and gets its own browser so it can carry out tasks across several apps and websites at once. You choose the model behind the agent: the AI provider can be set by you.

The example tasks from the announcement are everyday ones, which is why they are easy to grasp:

• prepare you for calls: gather context about the person you are speaking with and their company;
• collect invoices: go through service accounts and pull out the documents you need;
• cancel a subscription: work through menus deliberately designed to be inconvenient.

All three scenarios have one thing in common: there is no clean interface for a machine, only an interface for a human. An agent with a browser uses precisely that.

Why "its own browser" and "on your computer" matter

n8n users are accustomed to a different logic: you draw a chain of nodes where every step is known in advance. It is reliable, but brittle when a site changes its layout or a service has no integration. An agent works the other way around: you state a goal, and it works out the steps itself. The price of this approach is lower predictability.

Running locally gives two practical benefits. First, your data and sessions stay with you rather than going wholesale to someone else's cloud. A caveat: if you connect an external model provider, part of the context will still go to it. Second, it is easier for the agent to work with whatever you have access to, including services without a public API.

A separate browser for the agent is a deliberate decision. It does not poke around in your work tabs and does not get in your way while it carries out a task. How deeply this browser is isolated and how login storage works is not stated in the post.

What the announcement lacks

It is worth noting soberly that this is a teaser post, not a technical description. It does not include:

• a license and a repository with clear terms of use;
• a list of supported providers and local models;
• data on how reliably the agent completes real-world scenarios;
• a description of safeguards against mistakes: what happens if the agent cancels the wrong subscription.

The last point is the most important. An agent that can click on your behalf is equally useful and dangerous. Canceling a subscription is a relatively harmless action, but a payment, deleting data or an email to a client already requires human confirmation. Good agents are built around that question, not around a pretty demo.

What to do with this

If you are a developer or a founder, there is no need to rush to put Nod into your workflows. A sensible order looks like this:

1. Wait for the repository and read the license and the architecture description.
2. Try it on an isolated machine or a separate profile, with no access to payment details or work accounts.
3. Start with tasks where a mistake is cheap: collecting documents, preparing call notes.
4. Compare it with what you can already do in n8n with ordinary workflows. If the steps are stable, a deterministic workflow will be more reliable and cheaper than an agent.

For the market, the signal is more interesting than the product itself. A visual automation platform is releasing a local open-source agent with a choice of model, which means the line between "workflow" and "agent" is blurring, and the bet is on user control rather than a closed cloud. How well Nod survives contact with real websites will become clear once people start running it not in a demo, but in real life.