Short answer: On 10 November 2026 n8n Labs released Nod, a free, open-source (Apache 2.0) personal AI agent for Mac and Windows that runs in its own sandboxed computer and browser on your machine. It works in Gmail, Calendar, Slack and other apps and on websites with no integration, and asks before it sends, pays, posts or deletes. You bring the model: an API key, a ChatGPT Plus or Pro sign-in, or a local model. Costs are your model usage; Nod checks in when a task passes $5. It’s a research preview: one owner per install, it stops when your laptop sleeps, and you don’t need n8n to use it.
Checked 11 November 2026 (Asia/Tehran) against n8n’s announcement, the Nod website and docs, and the GitHub repository and releases. We haven’t run Nod ourselves.
n8n is best known for shared workflows that a team builds once and runs on a server. Nod, announced by Ophir Prusak on the n8n blog, goes the other way: one agent for one person, with its own computer. n8n calls it an n8n Labs experiment that “doesn’t change the direction of n8n”. Here’s what it does, what it needs, what it costs and where it fits next to n8n.
What n8n announced
| Item | Detail | Source |
|---|---|---|
| Release | Research preview from n8n Labs, 10 November 2026 | n8n blog |
| Platforms | macOS 14+ (Apple silicon or Intel), Windows 11 x64 with WSL 2; self-host on Linux or a Mac with Docker | install docs, self-host docs |
| Licence | Apache 2.0, code on GitHub | n8n-io/nod |
| Latest version | v0.6.1 (10 November 2026), after v0.6.0 on 9 November | release notes |
| Price | Free; you pay your model provider (or use a ChatGPT plan or a local model) | n8n blog |
| Channels | Desktop app, Telegram, WhatsApp | Nod docs |
| Apps | Gmail, Calendar, Drive, Slack, Notion, Linear and more; MCP servers; any website through its own browser | Nod docs |
| Needs n8n? | No; you can optionally connect your n8n instance | n8n blog |
The repository was created on 25 September 2026 and had 36 GitHub stars when we checked, so this is a very early public release.
What Nod does
Because Nod has a sandboxed Linux workspace and a browser of its own, it can finish jobs a chat assistant would hand back as a to-do list. n8n’s examples include signing in to supplier portals each month to download and file invoices, pulling last week’s numbers from a dashboard with no export into a weekly report, turning client requests from email and Slack into Linear tasks, chasing an unpaid invoice with a reminder (and asking before a second nudge), and cancelling a subscription while turning down the retention offers.
It runs several jobs in parallel, can run recurring jobs on a schedule (each schedule shows its estimated monthly cost) and keeps months of conversation in memory. You can see, edit or delete what it remembers. When you connect Gmail, it learns your writing style from sent emails.
What it costs
Nod itself is free. The bill is model usage, and n8n is candid that “some early testers found tasks cost more than they expected”; it says the latest release brought costs down “substantially”, and v0.6.1’s notes list cheaper background checks with background spend now shown separately. The built-in guardrails:
- A cost on every task and a receipt when it finishes
- A monthly spending limit you set
- A check-in once a task passes $5 (adjustable)
- An estimated monthly cost on each schedule
Model options from the models docs:
| Connection | How you pay | Note |
|---|---|---|
| OpenAI / ChatGPT | API key, or sign in with ChatGPT | ChatGPT sign-in needs Plus or Pro and uses the plan’s included usage |
| Anthropic | API key | Anthropic’s terms require an API connection for Claude in Nod |
| Google Gemini | AI Studio API key | Gemini 3.8 Flash has a free tier (quotas; different data-use terms); a Gemini app subscription can’t be used |
| OpenRouter | API key or browser sign-in | Shows per-model token prices |
| Bedrock, Vertex AI, Azure OpenAI, Cloudflare | Your cloud account | Bring your own region and deployment |
| Local (Ollama, LM Studio, vLLM, llama.cpp) | Free per request | Via an OpenAI-compatible server; see local models |
What you need to run it
- Mac: macOS 14 or later; the app and runtime are about 1.8 GB to download.
- Windows: Windows 11 x64 with WSL 2, hardware virtualization and at least 20 GB free disk; about 2.1 GB to download.
- Your own server: Linux AMD64 or ARM64, or a Mac with Docker Desktop. n8n suggests starting with 2 CPU cores, 8 GB of memory and 50 GB of disk, Docker Engine 25 or newer with Compose v2, and an SSH tunnel or a domain with ports 80 and 443 (self-host docs). That’s a bigger box than a small n8n instance; see our n8n RAM guide and VPS comparison for comparable server prices.
A laptop install stops when the laptop sleeps. n8n says some testers keep Nod on an always-on Mac mini or a server, and a hosted version is on the roadmap.
The trust model, and its limits
Nod is built to be checked rather than trusted blindly. A server-side trust check compares any send, payment, post or delete with what you actually asked for, so a web page or email can’t simply give it orders. Saved logins and cards are encrypted on your machine and entered through private forms the model can’t read, and every sign-in, card use and app action is logged with its cost.
n8n also lists the gaps. The permission check is itself a model and can be wrong. Its comparison page notes that Nod stores a card in its vault and can’t enforce the amount at the card issuer (unlike single-use cards), that its server can read a saved value while using it, and that there’s no published bug bounty. Managed app connections pass through an n8n-hosted connection service and Composio, and basic usage analytics are on by default for new owners (you can turn them off; conversation traces are only shared if you opt in).
n8n’s own test numbers (with GPT-6.1 Sol): 98% task completion across 226 errands, each run three times, with no actions taken without permission; 93.2% on the LongMemEval-S memory benchmark (500 questions); and 0 of 216 false memories planted by 108 published attack emails on WhisperBench (Nod website). These are vendor-run results.
What it means if you automate with n8n
| Nod | n8n workflow / n8n AI agent | |
|---|---|---|
| Built for | One person’s ongoing jobs | Shared team and business processes |
| Runs on | Your computer or your own server, in a sandbox with a browser | Your n8n instance or n8n Cloud |
| Works with sites without an API | Yes, through its own browser | Through integrations, HTTP requests or community nodes |
| Users | One owner per installation (sharing is “next”) | Many users, one instance |
| Cost | Free app + model usage | n8n pricing: per-execution plans on Cloud, or self-hosted |
| Maturity | Research preview, v0.6.1 | Production platform |
Our take: Nod is a useful companion, not a replacement. Keep anything a team depends on (lead routing, invoicing syncs, CRM updates) in n8n workflows, where runs are versioned, monitored and shared; that’s where our n8n workflow auditor and n8n cost calculator help. Nod fits the long tail of one-person admin that has no API: portal downloads, dashboards without exports, form-filling. Before you hand it a mailbox or a card, set the monthly limit, start with read-only jobs and watch the Usage page for a week. For context on other agents with their own accounts and memory, see Microsoft Copilot Studio’s persistent agents; for AI tool prices, see our AI SaaS pricing comparison.
What we couldn’t verify
- Real cost per task. n8n publishes no typical cost figures; it depends on your model and job. We didn’t run Nod.
- Benchmarks. The 98%, 93.2% and 0-of-216 results are n8n’s own tests on one model.
- Data path details. Exactly what the n8n-hosted connection service and Composio see for each app wasn’t checked beyond the docs’ summary.
- Hosted version and team sharing. On the roadmap, with no date or price.
- How the n8n connection works in practice. n8n says you can connect your instance; we didn’t test which workflow actions Nod can trigger.
- Linux desktop app. Linux is covered by source install and self-hosting; we didn’t see a Linux desktop installer in the release assets.
Frequently asked questions
What is Nod from n8n?
Nod is a personal AI agent released by n8n Labs on 10 November 2026 as a research preview. It’s a standalone desktop app for Mac and Windows, separate from n8n, that has its own sandboxed computer and browser. You give it a job, such as downloading invoices from supplier portals or chasing an unpaid invoice, and it works in your connected apps and on websites until the job is done, asking before it sends, pays, posts or deletes anything you didn’t ask for.
Is Nod free?
The app is free and open source under Apache 2.0. You pay your AI model provider for usage, sign in with a ChatGPT Plus or Pro plan, or run a local model through an OpenAI-compatible server such as Ollama. n8n says some early testers found tasks cost more than expected; Nod shows a cost receipt per task, supports a monthly spending limit and checks in when a task passes $5.
Do I need n8n to use Nod?
No. n8n says you don’t need an n8n account. You install Nod on your computer, connect your accounts and a model, and start. If you do run n8n, you can connect your instance so Nod can work with your workflows.
What are Nod’s system requirements?
The desktop app needs macOS 14 or later (Apple silicon or Intel, about 1.8 GB download) or Windows 11 x64 with WSL 2, hardware virtualization and at least 20 GB free disk (about 2.1 GB download). A self-hosted server needs Linux AMD64 or ARM64 or a Mac with Docker; n8n suggests starting with 2 CPU cores, 8 GB of memory and 50 GB of disk, Docker Engine 25 or newer with Compose v2.
Is Nod safe to give access to my email and cards?
Nod is designed with a server-side permission check that compares each action with your own words, a vault that encrypts saved logins and cards and enters them through forms the model can’t read, and a log of every sign-in, card use and app action. n8n says the permission check is itself a model and can make mistakes, and its own comparison page notes that Nod can’t enforce a payment amount at the card issuer. Treat it as a research preview.
How is Nod different from an n8n workflow or the n8n AI agent?
n8n runs shared, repeatable processes for a team on a shared instance. Nod is a single-player agent for one person’s ongoing jobs, including websites with no integration, using its own browser. Each Nod installation has one owner, it stops when the laptop sleeps unless you run it on an always-on machine or server, and it can’t make phone calls yet.
Related reading
- I connected Gemini’s free API to MausBot: 3.8-Flash worked, 2.5-Flash 404’d for new users
- Customers are sending AI agents to your support desk: Zendesk, Intercom and NiCE
- New AI agent tools this week (13 November 2026)
- New AI agent tools this week (6 November 2026)
- Microsoft Copilot Studio persistent agents: what Copilot Credits cost
- n8n 3.0 breaking changes
- n8n Cloud vs self-hosted
- How to secure your n8n instance
- ZCode vs Manus: AI agents compared
Sources
n8n blog: Meet Nod (10 Nov 2026); Nod website; Nod docs: overview, models, local models, install the desktop app, self-host, how Nod compares; GitHub repository; v0.6.1 release notes; n8n pricing. All checked 11 November 2026 (Asia/Tehran).
