by n8n Labs
Open-source personal agent that works in your apps and browser
Pick Nod if you run your own work and want an agent on your own machine that cancels trials, chases invoices and preps calls, with approvals and a spending limit. Skip it if you need a hosted assistant that keeps going while your laptop sleeps. The vendor's own benchmark page lists 226 test errands.
It runs on your own computer with its own sandbox and browser, and the limits on what it may do are enforced in a server check you can read in the open code.
Open code and a local-first design earn trust from privacy-minded buyers, but anyone can fork Apache-licensed code, so the lasting edge is the reach of n8n's brand and community.
Main trade-off
Unattended schedules depend on a machine that is always on and awake.
Free to download and open source, with no Nod subscription listed by the vendor. The running cost is whatever your AI provider charges, or nothing extra if you use a plan you already hold.
Asks before it spends
| System | Type |
|---|---|
| Gmail | Native |
| Slack | Native |
| Notion | Native |
| Google Drive | Native |
| Google Calendar | Native |
| Telegram | Native |
| Native | |
| n8n | API |
Nod is an open-source personal agent from n8n Labs, released as a research preview on 10 November 2026. It runs on macOS and Windows with its own sandbox and browser, works in connected apps and on websites, and uses the AI provider you choose. A server-side check asks you before any action your request does not cover.
Own sandbox and browser
Works in a local sandbox with a browser of its own, so it can finish jobs on websites that have no integration.
Connected apps and chat channels
Connects to Gmail, Slack, Notion, Google Drive and Calendar, and takes requests from the app, Telegram or WhatsApp in one conversation.
Approval check outside the model
A server-side check compares each action with your request and asks when it is not covered; every payment waits for your approval.
Vault for logins and cards
Saved logins and cards are encrypted on your machine and entered through private forms, so the model never receives them.
Memory you can edit
Remembers months of conversation, scoring 93.2% on LongMemEval-S in the vendor's own run (its benchmarks docs page), and lets you view, edit or forget each stored note.
Schedules and proactive jobs
Runs repeating jobs such as a morning brief on a schedule, and reads connected apps to suggest jobs it could take off your plate.
Spend tracking
Shows what each task cost, ends it with a receipt, checks in once a task passes $5 by default and supports a monthly spending limit.
Open code and self-hosting
The full code is on GitHub under Apache 2.0, and it can run on a server for one person instead of a laptop.
| Agent | Where it runs | Price | Maturity | Rating |
|---|---|---|---|---|
| Your machine | Free | Research | 3.3 | |
| Cloud · Your machine · API | Free | Beta | -- | |
| Cloud · Your machine · API | Free plan | Available | 3.7 | |
| Cloud | Free plan | Available | -- |
Nod itself is free to download and open source, and the vendor lists no Nod subscription. You pay your AI provider for the requests Nod makes with your API key, or you use a ChatGPT Plus or Pro plan you already have. Nod shows each task's estimated cost in Usage and supports a monthly limit.
The app is free but needs a model connection to work. The vendor's docs say Gemini 3.8 Flash has a free tier through a Google AI Studio key, subject to quotas and different data-use terms, and a model running on your own computer costs only your hardware. A ChatGPT Plus or Pro sign-in also works.
Nod stores its database, files, memory and saved logins on the computer or server that runs it. The model service you pick, your connected apps and the websites involved receive the data each task needs, and app connections go through n8n and Composio by default. Usage data goes to n8n unless you turn it off in Settings, and the vendor says it is stored in an EU project.
Nod connects to OpenAI with an API key or ChatGPT sign-in, Anthropic with an API key, Google Gemini and Vertex AI, OpenRouter, Amazon Bedrock, Azure OpenAI, Cloudflare and many other providers, or to a local model server. A Gemini app subscription is not a sign-in option, and the vendor says Anthropic's terms require an API connection for Claude.
Reading your inbox or calendar and drafting replies run without asking. Sending an email, adding an event or deleting files runs after a review checks that it matches your request, and Nod asks you when it does not. Every payment waits for your approval with your card's security code, which Nod never saves, and the vendor says the check is a model and can make mistakes.
Nod is a personal agent from n8n Labs, an experimental team inside the company behind the n8n workflow tool. It launched as a research preview on 10 November 2026 as a standalone app for macOS and Windows, separate from n8n and from the AI Assistant inside it, and it needs no n8n account. You describe a job in the app, on Telegram or on WhatsApp, and it works in your connected apps and on websites until the job is done: cancelling a trial, chasing an invoice, pulling statements from a supplier portal each month.
Nod runs its own sandbox, a private virtual machine on your computer with its own browser, and plans each job with the model you connect. The vendor's docs name GPT-6.1 Sol, Claude Opus 5.5 and Gemini 4 Argon among the starting choices, and a model on your own machine also works. A separate check, itself a model, compares each send, payment or deletion with your own words and asks you when they do not cover it. Saved logins and cards are entered through private forms the model cannot read, and repeating jobs such as a morning brief or a price watch run on a schedule.
It fits people who run their own work: freelancers, solo operators and small teams with recurring admin such as client call briefs, month-end paperwork and refund requests. The vendor also describes household errands like booking a hotel once you say go.
It is the wrong pick if you want a hosted assistant that keeps working while your laptop sleeps, because Nod stops with the computer that runs it and the vendor lists a managed version only as a roadmap item. Setup also expects an API key or a ChatGPT Plus or Pro plan. If you want a chat-first agent on a vendor's cloud computer, Muse is closer, as is OpenAI dots if you already pay for ChatGPT Pro. For a self-hosted gateway into many chat apps, start with OpenClaw, with Hermes Agent as the second option. Teams that want scheduled cloud automations should look at Tasklet.
What Hokai saw in the public record on 11 November 2026: the GitHub repository held 1,052 commits and an Apache-2.0 LICENSE file, and release 0.6.2, published that morning, listed Mac Intel, Mac ARM and Windows installers plus a signature file. Three desktop releases landed on consecutive days, 0.6.0 on 9 November, 0.6.1 on 10 November and 0.6.2 on 11 November, covering cheaper background work, clearer model errors and onboarding changes. Browse its peers in the personal assistant agents list.
Verdict: install Nod if you run your own work, are comfortable adding an API key and can leave a computer on for scheduled jobs. Skip it for now if you want a hosted assistant, an agent shared by a team or phone calls, none of which it offers yet.
Autonomy, guardrails, integrations and pricing on this page are read from n8n Labs's own documentation. The verdict is Hokai's reading of them, not the vendor's claim.