Most AI tools wait for you to type. This one clocks in and does errands. On November 10, 2026, n8n Labs — a new team inside n8n, the company behind the automation platform we build on — released Nod, a personal AI agent that runs on your own computer and carries out real tasks in your real apps. It’s out now as a free research preview, and it’s genuinely different from both the chatbots you’re used to and the workflow platform n8n is known for.
A personal agent with its own computer
Nod is a standalone product — it is not n8n, and it’s not the AI Assistant inside n8n. You don’t need an n8n account to use it. What makes it different is the setup: Nod runs on your machine in a private sandbox, with its own browser. After a single sign-in, it can work in the apps you already use — Gmail, Slack, Notion, Google Drive, Calendar — and it can also operate real websites directly, including sites that have no integration at all, using your saved logins.
That last part matters. Most “AI assistants” stop at whatever has an official connector. A lot of small-business work lives in exactly the places that don’t: the vendor portal, the booking page, the dashboard your supplier built in 2014. An agent with its own browser can work there too.
How this differs from a chatbot — and from n8n
A chatbot answers questions. n8n workflows automate team processes: you design the steps, and the system runs them the same way every time. Nod is built for the space in between — the single-player errands that fill an owner’s day but were never worth designing a whole automation for. You describe the outcome, and Nod figures out the steps in your actual apps, then checks back when something needs your judgment.
What it actually does all day
The demo jobs n8n shows are refreshingly unglamorous — the real work of running a business:
- A one-page brief before every call — pulling together the email threads, the calendar context, and the client’s latest news before you walk in.
- Spotting unanswered questions in your inbox and drafting replies sourced from your own documents.
- Watching pages for changes — for example, a client’s pricing page — and alerting you when something moves.
- Chasing invoices, politely — and asking you before it sends a second nudge.
- Cancelling trials and requesting refunds before a forgotten subscription renews.
- Filing month-end paperwork, turning email and Slack requests into Linear tasks, and comparing travel options — then booking only once you approve.
The permission model — the part we care about most
An agent that can act in your accounts is only useful if it never acts beyond what you asked. Nod is permission-first: it checks each action against your request and your rules, and asks before doing anything those don’t cover. In n8n’s own evaluation, across 678 test runs, Nod took zero actions without permission, and it completed 98% of 226 test errands. Those are n8n’s numbers, from simulated accounts and websites in October 2026 (running GPT-6.1 Sol), so treat them as the maker’s claim, not an independent benchmark — but the design priority is exactly right, and it’s the standard any agent working in your business should be held to.
It remembers — and you can see what it remembers
Nod keeps notes as it learns how you work: your preferences, your people, your writing style (it learns your voice from your sent mail). Nothing is hidden — every note is visible, editable, and deletable, and each one shows the message it came from. On the LongMemEval-S memory benchmark, Nod scored 93.2%. For a small business, that combination — real memory, fully inspectable — is what makes an agent feel less like a tool you prompt and more like a staffer who actually knows the place.
What it costs, and how to try it
Nod is free and open source (Apache 2.0 — the code is on GitHub). You bring your own AI: it works with 30+ model providers (OpenAI, Anthropic, Google, OpenRouter, and more), a ChatGPT Plus or Pro plan, or a fully local model via Ollama, LM Studio, or vLLM (the model needs tool calling and at least a 64k context window). You pay only your own model costs, and Nod keeps you in control of them: a receipt for every task, a monthly spending cap, and a check-in whenever a task passes $5.
It runs on macOS 14+, Windows 11 (in preview), or your own server. One practical catch: your machine has to stay on and awake for scheduled work to run. Managed hosting, phone calls, and a shared team version are on n8n’s roadmap. Full details are on the Nod product page and in the launch announcement.
The honest caveats
This is a research preview, and you should expect rough edges. Independent reports from the first days describe a slow first run on a free model tier, and some early testers saw higher model costs than they expected — n8n says its latest release has already cut those costs. And while your data stays on your machine, your model provider still sees whatever a given task needs, so which model you plug in matters. None of that is a dealbreaker for trying it on low-stakes errands; all of it is a reason not to hand it your bookkeeping on day one.
Our take
We build business automations on n8n every week, so a personal agent from the same house naturally has our attention. But the reason we’re writing about Nod isn’t the logo — it’s the pattern. The agents that will earn a place in small businesses aren’t the ones that act the fastest. They’re the ones that show their work, keep their memory where you can see it, put a receipt on every task, and ask before they act on your behalf. That’s the standard we build to at Launch AI Works, and it’s good to see it shipping in a free tool any owner can try this week.