Insights

What is Nod? n8n Labs' open-source personal agent, explained

As of 18 November 2026. Nod is a few weeks old, and personal agents are changing from one release to the next. Every fact below comes from n8n's own pages, checked on that date and listed at the end.

A personal agent does errands rather than answering questions: it signs in, fills in the form, sends the follow-up. Most of the agents launched this year run in the vendor's cloud, on the vendor's model. Nod takes the other road — open source, on your own machine, with the model you choose. Here is what it is, how it works, and who should wait.

What Nod is, and who makes it

n8n released Nod as a research preview on 10 November 2026. It comes from n8n Labs, a small team inside n8n that tries ideas outside the core product, and the site is candid that Nod is an experiment and that most Labs ideas get shelved. Nod describes itself as an open-source personal agent "with its own computer": a sandbox and a browser running on your machine, where it works through apps and websites until a job is done.

Two things to keep apart. Nod is not the n8n AI Assistant, and it does not need n8n — no account, no instance. If you already run n8n, you can connect your instance so Nod can use your workflows. Releases are frequent: the latest on GitHub when we checked was v0.8.4, published on 17 November 2026.

Who Nod is for

n8n built it for one person and that person's accounts, for work or for the household. The examples are deliberately ordinary: chasing an unpaid invoice, signing in to supplier portals each month and filing the invoices for the accountant, preparing a one-page brief before a client call, cancelling a subscription and asking for the refund.

For now you connect your own AI provider with an API key or a ChatGPT plan, and n8n itself says that suits people comfortable with that kind of setup. In a European SMB, that is the owner, a freelancer, or the one person who already looks after the tools — not a company-wide rollout.

How it works: its own computer, approvals and the Vault

You talk to Nod in one conversation — in the desktop app, on Telegram or on WhatsApp — while it runs several jobs in the background. Recurring jobs, such as a morning brief or a price check, run on a schedule. It works in connected apps such as Gmail, Calendar, Drive and Slack, on websites that have no integration at all, and with MCP servers, APIs and command-line tools.

The permission model is the interesting part. Reading and drafting happen freely. Before Nod sends, pays, posts or deletes anything, a separate check confirms that your request actually covers that action; when your instructions do not cover it, Nod asks you. Web pages, emails and files are treated as data, so a message in your inbox cannot give it orders. n8n states that these rules are enforced by Nod's server, not by the model's good behaviour, and every sign-in, card use and app action is logged together with its cost.

Saved logins and cards live in the Vault, encrypted on your machine. The model chooses the account and the field; the server fills in the value, so the model never sees the secret. n8n is just as clear about the limits: the permission check is itself a model and can be wrong, a website keeps whatever you let Nod submit, and the Nod server can read saved values while it uses them.

Licence and cost

Nod is free and released under the Apache-2.0 licence — including the permission check and the Vault. You pay your model provider directly, or nothing extra if the model runs on your own hardware. The cost controls are practical:

  • a receipt at the end of every task, showing what it cost;
  • a monthly spending limit, set to $200 when you finish setup and adjustable;
  • a check-in once a task passes $5, an amount you can change;
  • an estimated monthly cost for each schedule.

n8n acknowledges that early testers found tasks cost more than they expected, and says recent releases brought that down substantially.

Which AI models you can use

Nod connects to more than 30 providers. The named connections include Google Gemini, OpenAI/ChatGPT, Anthropic, OpenRouter, Google Vertex AI, Amazon Bedrock, Azure OpenAI and Cloudflare Workers AI, plus any OpenAI-compatible server. Further down the list are DeepSeek, Mistral, Moonshot AI, xAI, Groq, Together AI, Hugging Face, NVIDIA and others.

A few details matter in practice. A ChatGPT Plus or Pro plan works by signing in. Claude needs an Anthropic API key, because Anthropic's terms require an API connection for Claude in Nod. Gemini needs an AI Studio API key; a Gemini app subscription is not a sign-in option. For OpenAI, a connection can be set to OpenAI's Europe (EEA + Switzerland) region for projects with EU data residency. And Mistral, the French provider, is on the list for anyone who prefers a European model company.

Local models run through Ollama, LM Studio, vLLM or llama.cpp. n8n's guidance is that a local model needs tool calling and a 64,000-token context, and that small models are less reliable at the safety check. You can keep several connections, switch at any time, and give schedules or background work a lighter, cheaper model. If a request fails, Nod does not quietly move to another connection: it shows you the provider's reason and lets you choose.

Where it runs

The desktop app supports macOS 14 or later; Windows 11 and Linux x64 (as a Flatpak) are in preview. You can also build it from source on macOS or Linux with Node.js 24.21 or later, pnpm and Docker, or host it on your own server for one person. A managed hosted version is on the roadmap, not available today.

The practical constraint is uptime. Nod needs the machine running it to stay on and awake; on a laptop, it stops when the laptop sleeps. To keep schedules running, n8n suggests a Mac mini that stays on or your own server — which, for a European business, can be a server or VPS in a European data centre you already use.

Where your data lives, and the defaults

Nod stores its database, files, memory and saved logins on the computer or server that runs it. That matters, but it is not the whole picture: the model provider, the apps and the websites receive what each task needs. Run Nod in Lisbon with a US model provider and the task content still goes to that provider. Here, the choice of model is the sovereignty decision, not the location of the app.

Three defaults are worth knowing. App connections use n8n and Composio by default. And usage analytics are on by default for new owners; you can turn them off in Settings → Privacy & data. Conversation traces stay off unless you turn them on. Nod's memory spans months of conversation and learns your preferences, and you can see, edit or delete every note it keeps.

Maturity and honest limits

n8n publishes its own test figures: across 678 test runs, Nod took zero actions without permission, and 98% of jobs finished; on LongMemEval, a public long-term-memory benchmark, it scored 93.2%. These are n8n's own numbers and have not been independently verified.

The limits are stated plainly too. Each installation has one owner. Nod cannot make phone calls; they are listed as coming next. Nothing runs while the machine sleeps. And this is a research preview from a Labs team, not a supported product with a contract behind it.

Who should wait

Wait if you need one agent shared by a team, a vendor contract and support, or a deployment that someone else maintains. Wait, too, if you need certainty about where every piece of data goes, because the model provider still sees each task. If you already manage your own API keys and are happy to test Nod on errands that can go wrong without consequence, it is worth an afternoon now. And if you would like help deciding where an agent like this fits — and on which model — tell us what needs solving.

Sources (checked 18 November 2026)

Frequently asked questions

Is Nod the same as the n8n AI Assistant?
No. Nod is a standalone app from n8n Labs, a small team inside n8n. It does not need an n8n account or instance, although you can connect your n8n instance so Nod can use your workflows.
Is Nod free?
Yes. Nod is open source under the Apache-2.0 licence and free to download. You pay your model provider for the AI it uses, or nothing extra if you run a local model on your own hardware.
Can Nod run on a European model such as Mistral?
Yes. Mistral is one of the 30-plus providers Nod connects to, and Nod can also use local models through Ollama, LM Studio, vLLM or llama.cpp. Whichever model you choose receives the information each task needs.
Does Nod send data to n8n?
Usage analytics are on by default for new owners and can be turned off in Settings → Privacy & data. Conversation traces stay off unless you turn them on. App connections use n8n and Composio by default.
Can a whole team share one Nod?
Not yet. Each installation has one owner. A shared Nod for teams is listed on n8n's site as coming next, not available today.
← All insights

Let’s solve it.

Tell us about your business and what you’d like solved. We reply personally within one business day.

Start a project