Nod by n8n Tutorial: From Download to a Finished First Task
This tutorial takes you from zero to a working Nod setup: install the app, connect a model, connect your apps, set the safety and spending guardrails, and finish a real first task. Budget about 10 minutes for setup itself, plus a few minutes for the sign-ins. By the end you'll understand not just which buttons to press, but how Nod decides what it's allowed to do on your behalf.
First, the one-paragraph version of what you're setting up. Nod is an open-source personal agent from n8n Labs with its own computer — a sandboxed workspace and browser on your machine — so instead of advising you, it does the errand: chasing invoices, prepping call briefs, cancelling trials on real websites. It runs on a model you connect, and it asks before it takes actions you didn't request. It's a standalone app, not part of n8n; you don't need an n8n account. For the full picture, see What is Nod.
Step 1: Install Nod
The desktop app is the fast path — no Node.js, Docker, or developer tools required.
- Check your platform: macOS 14 or later (Apple silicon or Intel), Windows 11 (currently a preview build), or Linux x64 via Flatpak (also preview).
- Download the installer for your platform from nod.n8n.io or the
n8n-io/nodGitHub releases page — on a Mac, pick the DMG that matches your chip. - Install and open the app. On the Welcome screen, choose Get started to run the local assistant on this machine. (Choose Connect to a server only if you already run Nod on a server.)
- Wait for the runtime download to finish — on a Mac, the app plus runtime is about 1.8 GB.
Platform-specific walkthroughs with troubleshooting: Install on Mac and Install on Windows.
Step 2: Connect a model
Nod ships without a brain; everything it does runs on a model you connect in Settings → Models. You can add several connections and even use different models for different jobs, but you need one to start.
- Open Settings → Models and add a connection.
- Pick your route:
- ChatGPT Plus or Pro sign-in — if you already pay for ChatGPT, sign in with that account and use it as the model source. No separate API purchase needed. This is the fastest route for most people.
- An API key — OpenAI, Anthropic, Google Gemini (via an AI Studio key), OpenRouter, DeepSeek, Groq, Mistral, xAI, and many more; the docs count 30+ providers, plus Vertex AI, Bedrock, Azure OpenAI, and any OpenAI-compatible endpoint.
- A local model — possible via tools like Ollama, but it has real requirements (tool calling and at least 64K context). Save this for later unless it's specifically what you came for.
- Select the model for the connection and save.
Two exceptions worth knowing, both from the official docs: Claude must be connected with an API key — a Claude subscription can't be used as a sign-in. And a Gemini app subscription can't be reused either, because Google doesn't allow third-party assistants to use it; Gemini needs an API key.
One piece of buying advice: pick a capable model for your main connection. Nod's safety checks are themselves model judgements, and in n8n's own published testing, its strongest models (GPT-6.1 Sol and Claude Opus 5.5) judged all 240 test decisions correctly, while a small model (Claude Haiku 5.5) let 8 unsafe actions through. Nod's own docs put it bluntly: pick a capable model. The full model-by-model walkthrough is in Connect a model.
Step 3: Connect your apps
A model gives Nod a brain; connected apps give it something to work on.
- Start with email and calendar. Together they unlock the flagship jobs: morning briefs before your calls, and spotting the customer question you forgot to answer.
- Add chat and task tools you live in (Slack, Linear, and the like) as you find jobs for them.
- Don't worry about services with no integration. Nod has its own browser and can work on ordinary websites using logins you save — the credentials are stored encrypted in its Vault on your machine and are filled in through private forms the model itself can't read.
One transparency note from the docs: app connections by default run through a connection service hosted by n8n and Composio. Your data, memory, and saved logins live on the machine running Nod. Details in Connect apps.
Step 4: Set the guardrails before you delegate
Do this step before your first real task, not after.
Understand the ask-first rule. Before Nod sends, pays, posts, or deletes anything your request didn't cover, an independent check compares the action against what you asked for and the rules you've set. If it isn't covered, Nod stops and asks you. Content from web pages and emails is treated as data, never as instructions — a page can't talk Nod into doing something.
Set a monthly spending limit. With an API model, the provider bills you directly for usage, and costs vary by task. Nod shows each task's actual cost and gives you a receipt when it finishes; any single task heading past $5 pauses mid-run to check with you (that threshold is adjustable). A monthly cap stops API-model work when it's reached. More on the money side: Pricing.
Check the privacy toggle. New installs share basic usage data with n8n by default. If you prefer not to, switch off Share usage data in Settings → Privacy & data. Conversation traces are only ever sent if you turn them on separately.
Know where the record lives. Every login, card use, and in-app action is logged in Usage, alongside each task's cost — when something looks wrong, that's your audit trail.
Step 5: Run your first task
Pick something real but observable. The morning brief is the ideal first run:
- In the app, ask in plain language: "Prepare a one-page brief for each of tomorrow's calls — read my calendar, the relevant email threads, and check each client's latest news."
- Watch it work. Nod reads your connected apps, browses in its own browser, and assembles the briefs — you can follow along rather than staring at a spinner.
- Respond to any confirmation prompts. On a first task, expect Nod to check in before anything outward-facing happens. That's the trust check from Step 4 doing its job.
- Review the result and the receipt: what it did, what it cost. Skim the Usage log to see the trail it left.
Other good first tasks, all from n8n's own list of examples: chase an overdue invoice (it sends a friendly reminder first and asks before chasing a second time), or set a price watch on a page (it checks every few hours and only notifies you when something changes).
In a hurry? Getting started is this same path compressed into the shortest possible version.
Step 6: Teach it, then put it on a schedule
Two features turn Nod from a novelty into a habit:
Memory. Tell it facts once — "Anna now handles invoices" — and it remembers. Every memory is visible, editable, and deletable, and labelled with the message it came from. It also learns your writing style from your sent mail, so drafts sound like you, and it remembers the steps of jobs it has done so it can repeat them.
Schedules. Recurring work — the morning brief, price checks, a monthly round of downloading invoices from vendor portals — can run on a schedule instead of waiting for you to ask. Each schedule shows an estimated monthly cost before you commit to it. Nod will also start suggesting jobs on its own: it reads your connected apps and proposes tasks it could take over, drafting the work and waiting for your approval before anything is sent.
You can also talk to it from Telegram or WhatsApp, and the same conversation continues across the app and your phone. Multiple tasks can run in the background in parallel.
Limits to keep in mind
- Sleep stops it. On a laptop, Nod works only while the computer is awake and the app is open. For an always-on agent, run it on an always-on Mac or your own server — managed hosting is on n8n's roadmap, not yet available.
- It can't make phone calls. That's listed as coming next, not current.
- It can be wrong. The safety check is a model's judgement and can err — which is exactly why the ask-first rule exists. Keep it switched on for anything that spends money or talks to other people.
More answers in the FAQ, and every guide in the series lives on the Nod hub.