Install Nod on a Mac
- Under Apple menu → About This Mac, check whether it says Chip (Apple silicon) or Processor (Intel).
- Download the matching DMG from the GitHub releases:
nod-arm64.dmg for Apple silicon, nod-x64.dmg for Intel.
- Open the DMG, drag Nod to Applications and open it from there.
- The first time, macOS asks whether to open an app downloaded from the internet. Select Open.
- On the welcome screen, choose Get started for an assistant on this computer, or Connect to a server if you already run a Nod server.
- If macOS asks for your keychain password, approve it. According to the docs, that's how Nod protects your saved keys.
Instead of the DMG you can use one Terminal command, according to the docs. Quit Nod first and run it as your normal user, without sudo:
(
installer="$(mktemp)" || exit
trap 'rm -f "$installer"' EXIT
curl -q --proto '=https' --tlsv1.2 -fsSL https://raw.githubusercontent.com/n8n-io/nod/main/deployment/n8n/install.sh -o "$installer" && /bin/sh "$installer"
)
According to the docs, the script finds the latest release on GitHub, verifies the downloaded app, installs it to /Applications/Nod.app and opens it. You don't need a GitHub account, and existing data stays in place. The command fetches the script itself straight from the main branch on GitHub, not from a fixed release. If you'd rather not, use the DMG.
Install Nod on Windows (preview)
- Download
nod-Setup-x64.exe from the GitHub releases.
- Run the installer. According to the docs, Windows shows n8n GmbH as the publisher; SmartScreen can still warn about a new release.
- Choose Get started or Connect to a server.
- If Nod needs WSL, follow its instructions and approve the administrator request. If Windows needs a restart, open Nod again afterwards; your connections and settings are kept, according to the docs.
Alternatively, quit Nod and install it from a normal PowerShell window (Windows PowerShell 5.1 or PowerShell 7), not as administrator:
irm https://raw.githubusercontent.com/n8n-io/nod/main/deployment/n8n/install.ps1 | iex
This script also comes from the main branch; the installer from the releases is the more cautious route.
Install Nod on Linux (Flatpak preview)
According to the docs, Linux only has a Flatpak preview for x64, no ARM builds. Because GitHub limits each release file to 2 GiB, the bundle comes in numbered parts.
Install Flatpak with your package manager and add Flathub:
flatpak remote-add --user --if-not-exists flathub https://dl.flathub.org/repo/flathub.flatpakrepo
Open Nod's releases, pick a release with Linux files and download every nod-linux-x64.flatpak.partNN file plus SHA256SUMS-linux-x64 into one empty folder.
In that folder, join the parts, check them and install the bundle. The last step only runs if every checksum passes:
cat nod-linux-x64.flatpak.part[0-9][0-9] > nod-linux-x64.flatpak && \
sha256sum --check SHA256SUMS-linux-x64 && \
flatpak install --user ./nod-linux-x64.flatpak
Open Nod from your application menu or run flatpak run io.n8n.Nod.
Keep your secret store unlocked during setup. According to the docs, Nod refuses to save keys without secure storage.
First launch: setup
Right after the welcome screen, Nod starts preparing its own computer and shows the progress. Keep your internet connection on. While that runs, setup takes you through these steps, according to the docs:
- Connect email (optional). Gmail or Outlook through Composio, the service Nod uses to connect apps. You can skip this and connect email later under Apps.
- About you (optional). A few questions about where Nod should help and what you want off your plate.
- Connect a model, if none is set up yet. More on that below.
- Conversation traces. You decide whether Nod sends traces to n8n; without your yes they stay off. The step also says that basic usage data goes to n8n by default. You can turn that off under Settings → Privacy & data.
- Email updates (optional), with its own consent.
At the end, Nod tries to turn on Open at login, so it's ready when you sign in to your computer. After the first greeting in chat you can allow notifications. How long Nod keeps working depends on the computer:
If tasks should also run while your laptop is closed, Nod belongs on a computer that stays on. How to do that is in Nod Self-Hosted: AI Agent by n8n on Mac mini.
Connect a model
Without a model, Nod can't think. You add connections during setup or later under Settings → Models, with Add connection. The model you choose decides your costs and how reliably the approval check works.
ChatGPT Plus or Pro by signing in
- Choose OpenAI, then Sign in with ChatGPT.
- Sign in to ChatGPT in the browser tab that opens and allow Nod to use your plan.
- Return to Nod. If the last page in the browser doesn't load, copy its full address into Callback URL and select Complete sign-in.
According to the docs, sign-in needs ChatGPT Plus or Pro; on other plans it fails. In a workspace your organization manages, an admin has to allow third-party apps first. ChatGPT then lists Nod under Settings → Usage, where you can see and limit its usage or disconnect it. Sign-in uses port 1455 on the computer that runs Nod.
Claude with an API key only
You can't sign in to Nod with a Claude subscription. The docs say: “Anthropic's authentication conditions also require an API connection for Claude in Nod.” This is how you connect Claude:
- Create an API key in Anthropic's console.
- In Nod, choose Anthropic, paste the key into API key and select Connect.
- If the key works for several workspaces in your organization, choose the right workspace.
You then pay Anthropic per use, on top of any subscription you already have. According to the docs, the initial choice is Claude Opus 5.5 at high effort.
Other providers and local models
According to n8n, more than 30 providers work, including Google Gemini, OpenRouter, Amazon Bedrock, Google Vertex AI and Azure OpenAI, plus any server with an OpenAI-compatible interface. According to the docs, Gemini 3.8 Flash has a free tier with quotas; different terms apply there to what Google does with your data. Local models run through Ollama, LM Studio, vLLM or llama.cpp and, according to n8n, need tool calling and a 64,000-token context. With Nod, choosing a model is also a safety decision, because the approval check is itself a model. What that means is in the section on models and costs in Nod by n8n: What the AI Agent Can Do.
Set a monthly limit
With an API model, Nod shows what each request cost under Usage (⌘5 in the Mac app). According to the docs, Nod sets a monthly limit of 200 US dollars when you finish setup. To change it:
- Select an API model and open Usage.
- In the sentence at the top, click your current limit.
- Enter an amount in US dollars, from 1 dollar up, and select Save.
You can also tell Nod in chat: “Set my monthly limit to $40”. When the limit is reached, schedules, tasks and background work stop before Nod calls the model. Work that's already running can finish, and an action you already approved still runs once. Nod then only answers your messages to change the limit. On top of that, according to the docs, every task pauses once it has spent 5 dollars since your last message to it and asks whether to keep going.
With a ChatGPT plan, Usage hides costs and the limit. Plan usage doesn't count toward the monthly limit, according to the docs, but the 5-dollar check-in still counts it at API prices. The amounts are Nod's estimates; your provider's bill is the final word. What each model option costs per month, with worked examples, is in Nod Pricing: What the AI Agent by n8n Costs.
Your first task
After setup, Nod opens the chat with a short greeting and idea cards: one task to do once and two that repeat. If you connected email, the cards are based on a small sample of it; otherwise they are general suggestions. A card is not your permission: according to the docs, its words are Nod's, not yours, and they can't approve sending or deleting anything.
You can start without connected apps. The docs suggest pasting three supplier quotes and asking Nod to compare prices and delivery times, with your deadline and a recommendation. For a good first request, the docs give four rules:
- Say what done looks like, for example “a table with price and delivery time”.
- Name the account or site if Nod should sign in.
- Say whether Nod should send or only draft.
- Ask for schedules in plain words, for example “every Monday at 9 am”.
The square Stop button next to the send button stops Nod. According to the docs, it can't undo something that already started, such as an email already on its way.
Install from source
According to the docs, the source route is for people who want to change Nod's code. It runs on macOS, Linux and Windows. pnpm start installs dependencies, builds the web app, starts four Docker containers (PostgreSQL, the workspace, the workspace's firewall and the browser) and then the Nod server, which listens only on 127.0.0.1.
git clone https://github.com/n8n-io/nod.git
cd nod
corepack enable
pnpm start
The first run takes a while, according to the docs. When the server is ready, it prints a link such as http://localhost:4310/#access=…. Open it once; your browser then keeps the access, and http://localhost:4310 is enough. Then you go through the same setup as in the app.
The key Nod uses to encrypt saved credentials (the Vault key) is created automatically in the Keychain on macOS and in .local/vault/master-key on Linux. On Windows you create it yourself first and add it as NOD_VAULT_KEY to the .env file in the nod folder. Keep this key with your backup; without it, saved credentials can't be read.
On backups, the docs say themselves that there is no tested restore procedure yet. If Docker runs in a VM such as Colima, give it two CPUs and 8 GB of memory to start, according to the docs. Other devices can't reach this installation; for that you need the server option with a token and HTTPS. The desktop app doesn't import data from a source install, by the way; it starts with a fresh profile.
Where your data lives
With a local assistant, Nod's database, workspace and browser stay on your computer:
Requests go to the model provider and apps you choose. According to the docs, app connections go through a service n8n hosts and through Composio by default. Usage data goes to n8n by default; conversation traces only if you turn them on. What exactly goes where and how to turn off usage data is in Nod Privacy: What the AI Agent by n8n Shares.
Update Nod
On Mac and Windows, Nod checks for updates after it starts, every six hours and after your computer wakes, according to the docs. To check by hand, use Check for Updates… in the Nod menu or tray. When an update is downloaded, choose Restart to update. If a task is running, Nod asks before restarting. Profile, keys and data stay, and setup doesn't run again.
On Linux, update checks are off. You quit Nod and install the next bundle the same way as the first; according to the docs, flatpak update doesn't fetch a new Nod version. If your app is connected to a server, updating the app doesn't update the server.
Remove Nod
Moving the app to the Trash or uninstalling it keeps your data, so a later install picks up where you left off. To delete everything:
- Select Remove all data under Settings → Privacy & data. If Nod can't finish setup, use Remove all data… in the Nod menu.
- Read the warning and confirm. Nod deletes your profile and its own computer, restarts and opens setup.
- Then quit Nod and remove the app: to the Trash on a Mac, through Settings → Apps on Windows, or with
flatpak uninstall --delete-data io.n8n.Nod on Linux.
On Linux, removing only the Flatpak leaves Nod's own computer in ~/.nod-desktop; use Remove all data first. On a Mac, Nod keeps the runtime it downloaded and its key store entry, according to the docs, neither of which holds your data. Apps you connected may still list Nod until you remove it there.
Common errors
All cases come from the troubleshooting section of the docs:
For the failed Mac update, the docs give these two commands. They remove only the update job and its cache, not your profile:
launchctl bootout "gui/$(id -u)/dev.nod.desktop.ShipIt"
rm -rf "$HOME/Library/Caches/dev.nod.desktop.ShipIt"
If the first command reports an error because no job is loaded, continue with the second. One rule from the docs matters most: never delete Nod's profile or its own computer to fix a problem, or your conversations, workspace and saved keys are gone.
Frequently asked questions
Where do I download Nod?
From the releases of the official n8n-io/nod repository on GitHub, also linked from nod.n8n.io. There are DMG files for Apple silicon and Intel Macs, nod-Setup-x64.exe for Windows and a Flatpak in parts for Linux.
Do I need Docker or Node.js for Nod?
Not for the desktop app. You only need Docker, Node.js 24.21 and pnpm to install from source, and Docker for your own server.
Does Nod run on Windows 10?
The docs list Windows 11 x64 as the requirement, and Nod is still a preview there. For a local assistant you also need WSL 2 and at least 20 GB of free disk space.
Can I use my Claude subscription in Nod?
No. Claude only works in Nod with an API key from Anthropic, because of Anthropic's terms, according to the docs. A ChatGPT Plus or Pro plan, on the other hand, you can sign in with directly.
How long does installing Nod take?
n8n says setup with app, model and apps takes about ten minutes. On top of that comes the download of about 1.8 GB on a Mac or 2.1 GB on Windows.
Can I remove Nod with all its data?
Yes. First Remove all data under Settings → Privacy & data, then remove the app. Deleting only the app leaves your data in place.
What to do next
Install Nod without connected apps first and give it a task with material you paste into the chat yourself. That way you see how Nod asks and what a task costs under Usage before it gets near your inbox. Set the monthly limit to an amount you actually mean to spend. If Nod should run around the clock, the next step is your own server. The guide for that is Nod Self-Hosted: AI Agent by n8n on Mac mini. If you were actually looking for the workflow software n8n, see my guide to n8n with Docker and Kubernetes. How I build AI employees with clear permissions and approvals is what I show in my German-language community Claude Practitioners.
