The problem
How it works
n8n Labs released Nod as a research preview: a standalone app, separate from n8n, for Mac with Windows and Linux previews. You connect your accounts and a model provider, then give it jobs in its app or over Telegram or WhatsApp; it works in a local Linux sandbox with its own Chromium browser and can run several jobs in the background or on a schedule. n8n’s examples include cancelling a subscription and asking for a refund, watching a price, downloading monthly invoices from supplier portals for an accountant, copying figures from a dashboard that cannot export, and writing a one-page brief before each call. It remembers preferences, learns how you write from sent mail, and saves the steps of a task to reuse and repair when a site changes.
Results and limitations
Marked planned: the launch video, post and README describe the product and n8n’s own tests, not a user’s project. The README reports 98% of 226 simulated errands finished, no unsafe actions in 678 runs and 93.2% on the LongMemEval-S memory benchmark, measured in simulated accounts and sites.
Before you try it
Whether an action is covered by your request is decided by a separate check that is itself a model, so it can be wrong; n8n says Nod asks whenever your words do not cover an action. Saved logins and cards are entered through a vault the model cannot read, and pages and emails reach the model as data, not instructions. Set a monthly spending limit, since early testers found tasks cost more than expected. On a laptop it stops when the machine sleeps. Results depend on your setup and were not independently reproduced by Top 200 Repos.
Tools and repositories
No ranked repository is involved. This use case still stands on its own source.
- Tools
- Nod · Docker · A model provider or ChatGPT plan
- Prerequisites
- A Mac (or the Windows or Linux preview) that stays awake, and a ChatGPT Plus or Pro plan or a model API key.
Try it yourself: Hand over a recurring errand
Fill the fields to adapt the prompt. Empty fields stay as visible [placeholders].
Every [How often], sign in to [Site] with my saved login, download [Documents] for the last period, rename them as [File naming] and file them in [Folder]. If the page looks different from last time, or a step would pay, send or delete anything, stop and ask me first. Finish with a short receipt of what you did and what it cost.Run it once while watching, then check the files against the portal before you put it on a schedule.
Sources
- Introducing Nod: An Open-Source Personal AI Agent from n8n (opens in a new tab)
Vendor post · youtube.com · Published 11 Nov 2026 · Source checked 13 Nov 2026, 19:26 UTC
What this source supports
- Nod is an open-source personal agent from n8n Labs that runs on your computer
- Connects to apps and uses its own browser on websites without an integration; remembers preferences; supports scheduled tasks; provider of your choice
- Standalone research preview; no n8n account needed
- Meet Nod: a standalone personal agent from n8n Labs that runs on your computer (opens in a new tab)
Vendor post · blog.n8n.io · Published 10 Nov 2026 · Source checked 13 Nov 2026, 19:26 UTC
What this source supports
- Runs in a private local sandbox with its own browser; works in Gmail, Calendar, Drive and Slack and on sites without integrations
- Example jobs: cancel a subscription, watch a price, download supplier invoices, pull numbers from a dashboard without export
- A separate check confirms before it sends, pays, posts or deletes; the check is a model and can make mistakes
- Monthly spending limit and a check-in once a task passes $5; early testers found costs higher than expected
- Apache 2.0; stops when a laptop sleeps; one owner per installation; cannot make phone calls
- n8n-io/nod (opens in a new tab)
Project documentation · github.com · Undated · Source checked 13 Nov 2026, 19:26 UTC
What this source supports
- macOS 14+, with Windows 11 and Linux x64 previews; ChatGPT Plus or Pro or an API key
- Vault enters passwords, codes and cards so they stay out of the model; app and web results are marked as data
- Own tests: 98% of 226 errands finished over 3 runs each, 0 unsafe actions in 678 runs, 93.2% on LongMemEval-S
- One Node process plus Docker containers for the Linux workspace, Chromium and PostgreSQL