n8n Labs Releases Nod: A Local AI Agent with a Virtual Computer and No Account Required

There are already plenty of agents that click through websites on their own. Almost all of them require either a cloud subscription or registration in the vendor's ecosystem. n8n Labs took a different path. Their new project, Nod, has been released as an open research preview and runs on your own computer. No n8n account is needed.

For now this is a short announcement, and our coverage does not include detailed documentation. So below we separate the verified facts from what you should check yourself.

What is known today

Nod runs an AI agent inside a virtual computer on your machine and gives it a separate browser. The announcement names two scenarios: collecting invoices and tracking prices. Both will be familiar to anyone who has ever manually downloaded PDFs from a dozen customer portals or checked whether a needed service had gotten cheaper.

The requirements are as follows:

• macOS 14 or later;
• Windows is still in preview; it requires Windows 11 on x64 with WSL2 enabled;
• the program is free and open source;
• you have to connect a model yourself, and paid models are billed separately; Nod does not cover their cost.

There is also an honest limitation: if the computer goes to sleep, tasks are paused. This is a local agent, not a cloud worker that runs around the clock.

Why a virtual computer and a separate browser

The main pain point of such automations is that most routine processes have no API. An invoice from a hosting provider, a statement from a partner bank, or a price on a marketplace is available only through a web interface. Classic scripts built on Selenium or Playwright break with every layout change. An agent that looks at the screen and acts like a human survives such changes more easily.

But the approach has a second problem: security. If you let an agent into your main browser, it gets access to your sessions, email, and everything else. A separate browser inside a virtual machine limits the damage: the agent sees only what you have given it.

Imagine a small agency. At the start of the month, the accountant collects invoices from a dozen SaaS services, and the marketer checks competitors' prices once a week. This is an example, not a case from the announcement. If Nod handles such tasks, it fills exactly this niche: routine work that is a shame to hand to a developer and tedious to do by hand.

Limitations and open questions

First: this is a preview. It is too early to expect stability, accuracy, or protection against agent errors. Such a tool is not suitable for working with money and personal data without verification.

Second: you need your own model. Agentic scenarios, where the model looks at screenshots and makes decisions over and over, consume a lot of tokens. A "free app" can easily turn into a noticeable API bill. The alternative is a local model, but then you run into hardware and quality limits.

Third: the announcement doesn't make clear how Nod relates to n8n itself. Whether its tasks can be launched from workflows and how logins and passwords are stored has not been stated yet. These are the key questions, and the answers are worth looking for in the repository.

Fourth: "while the laptop is awake" is a serious practical barrier. For regular price monitoring, you will have to keep the machine on or dedicate a separate mini PC to the agent.

What to do in practice

If you are a developer or product manager eyeing computer-use agents, Nod is worth trying as a sandbox. Take one harmless task, such as tracking a price on a public page, and connect an inexpensive model. Measure how much a single run costs and how often the agent makes mistakes. Only after that should you think about invoices and private accounts.

Check right away where credentials are stored and whether the agent can be restricted to a list of domains. A separate browser on a virtual machine protects better than your main profile, but it doesn't replace basic hygiene. For users who care about privacy, Nod's key advantage is that everything runs locally and without mandatory registration.