n8n Nod: Setup, Local Models and Costs of the New Agent

Last updated: November 15, 2026

Nod is a free, open-source personal agent from n8n Labs, released as a research preview on November 10, 2026. It runs on your Mac, Windows PC, Linux machine or server, uses its own sandboxed computer and browser, works in apps such as Gmail and Slack, and asks before actions you did not request.

Contents

What Nod is
What changed this week
How to install Nod
Models and local setup
Cost and spending limits
Approvals and safety
Nod compared with other agents
Limitations
Frequently asked questions

What Nod is

n8n’s launch post describes Nod as a standalone app from n8n Labs, a new team that experiments with ideas outside the core product. It is separate from n8n, you do not need an n8n account,. For n8n’s own agent features, see our n8n 2.48 guide.

Nod is built for one person and finishes jobs that a chat assistant would hand back as a to-do list. n8n’s examples include:

Cancelling a subscription, turning down retention offers and asking for a refund.
Signing in to supplier portals each month and filing invoices.
Pulling numbers from a dashboard with no export and adding them to a weekly report.

You talk to Nod in one conversation, in the app or on Telegram or WhatsApp, while it runs several jobs in the background. Repeating jobs, such as a morning brief, can run on a schedule. n8n also says Nod remembers months of conversation and scored 93.2% on the LongMemEval long-term memory benchmark. (n8n’s own result).

What changed this week

Two releases since launch matter most for power users:

Release	Date (ET)	Main changes
0.7.0	November 12, 2026	Linux x64 Flatpak preview; local file work in the Mac, Windows and Linux apps, with folders you can keep off limits; new setups start with a $200 monthly model spending limit.
0.7.6	November 13, 2026	Automatic preparation of nudges and tasks from app changes, off by default for new profiles, with a separate adjustable $5 monthly allowance; Show commands for a task’s latest 200 workspace commands; commands in Usage and CSV exports; font size controls.

The 0.7.6 notes warn that command exports may still contain private file content.

How to install Nod

The desktop install guide lists these requirements:

Mac: macOS 14 or later, Apple silicon or Intel. The app and runtime are about 1.8 GB to download.
Windows: Windows 11 x64. A local assistant needs WSL 2 with hardware virtualization and at least 20 GB of free disk space. The download is about 2.1 GB.
Linux: x64 Flatpak preview with a desktop secret store. A local Linux assistant also needs KVM.

To set up a local assistant on a Mac:

Download the Apple silicon or Intel DMG from the Nod GitHub releases page.
Drag Nod to Applications, open it and approve the macOS prompt.
At Welcome, choose Get started. Choose Connect to a server instead if you already run a Nod server.
Connect a model service, or a local model server, when onboarding asks.
Choose whether to share traces. Basic usage data is shared by default and can be turned off in Settings.
Connect apps such as Gmail, Calendar, Slack, Notion or Drive, then give Nod its first job.

The Nod site estimates setup at about ten minutes. You can also run Nod from source or host it on your own server.

Models and local setup

Nod works with more than 30 providers, including OpenAI, Anthropic, Google and OpenRouter, or a ChatGPT Plus or Pro sign-in. The models page lists initial choices such as Claude Opus 5.5 at high effort for Anthropic and Gemini 3.8 Flash at medium effort for Google AI Studio. Claude requires an API connection, and a Gemini app subscription cannot be used to sign in.

For local models, the local models guide sets two firm requirements: the model must support tool calls, and the server must allow at least 64,000 tokens of context. Nod’s instructions and tools take about 20,000 tokens before your first message.

Server	Key setting	Address in Nod
Ollama	`OLLAMA_CONTEXT_LENGTH=65536 ollama serve`	`http://localhost:11434/v1`
LM Studio	Load the model with Context Length 65,536 or more	`http://localhost:1234/v1`
vLLM	`--max-model-len 65536 --enable-auto-tool-choice` plus a tool call parser	`http://localhost:8000/v1`
llama.cpp	`llama-server -c 65536 --jinja`	`http://localhost:8080/v1`

Then open Settings, Models, select Add connection and choose OpenAI-compatible. By default, the trust check also uses your local model at its lowest reasoning setting. You can pick a small model from another connection, or a decision service, as the action checker instead.

Cost and spending limits

Nod itself is free. You pay your model provider, or nothing per request with a local model. n8n says early testers found some tasks costly and that recent releases cut costs substantially. Guardrails:

Every task shows its cost and ends with a receipt.
A monthly spending limit, set to $200 for new setups since 0.7.0.
A check-in once a task passes $5, which you can change.
An estimated monthly cost for each schedule.

Approvals and safety

Before Nod sends, pays, posts or deletes anything you did not ask for, a separate check on the server confirms you asked for it. n8n says a web page or email cannot give Nod orders. Saved logins and cards are encrypted on your machine and entered through private forms the model cannot read, and Usage records every sign-in, card use and app action.

n8n notes the permission check is itself a model and can make mistakes, so Nod asks when your words do not cover an action.

Nod compared with other agents

n8n’s own comparison page says Nod is ahead on open source code, model choice and a hard spending limit. It also lists where Meta’s Muse is ahead: single-use card numbers tied to one merchant and amount, a bug bounty of up to $300,000, and placeholder tokens swapped at the network edge. Nod stores a card in its Vault and cannot enforce the amount at the card issuer. For those products, see our Muse Mail guide and ChatGPT Wallet guide.

For other personal agents, see our Alice AI Do it! guide and our Open Duet 0.8 guide.

Limitations

Research preview. n8n calls Nod an experiment.
Needs an awake computer. On a laptop, Nod stops when the laptop sleeps. A managed hosted version is on the roadmap.
One owner per installation. A team version is listed as coming next.
No phone calls. n8n lists calls as a future feature.

Frequently asked questions

What is n8n Nod?

Nod is a free, open-source personal agent from n8n Labs, released as a research preview on November 10, 2026. It runs on your own computer or server, works in your apps and on websites, and asks before actions you did not request.

Do I need an n8n account to use Nod?

No. n8n says Nod is a standalone app, separate from n8n and from the agents you build inside n8n, and you do not need an n8n account. It is not the n8n AI Assistant. You can still connect an n8n instance so Nod can use your workflows.

How much does Nod cost?

Nod is free to download under the Apache 2.0 license. You pay your model provider for usage, or connect a local model. New setups start with a $200 monthly model spending limit, and Nod checks in once a task passes $5. Both amounts can be changed.

Can Nod run with a local model?

Yes. Nod connects to Ollama, LM Studio, vLLM, llama.cpp or any OpenAI-compatible server. The model must support tool calls and at least 64,000 tokens of context, and n8n warns that small models are less reliable at the safety check.

Which computers does Nod run on?

Nod runs on macOS 14 or later (Apple silicon or Intel), on Windows 11 x64 as a preview that needs WSL 2 for a local assistant, on Linux x64 as a Flatpak preview, or on your own server for one person.

n8n Nod: Power-User Guide to the Open-Source Personal Agent