Overview
Nod is a personal agent from n8n Labs, a new team inside n8n, released on November 10, 2026 as an open-source research preview under Apache 2.0. It is a separate product from n8n: n8n runs processes a team shares on a shared instance, while Nod handles the jobs that belong to one person, installed on their computer with their accounts connected, no server and no n8n account required.
Because it has its own local virtual machine and browser, it goes one step past a chat assistant. A chat assistant hands the work back as a to-do list; Nod finishes it, operating inside connected accounts such as Gmail, Calendar, Drive and Slack and clicking through websites that offer no integration at all. Nod's own examples include cancelling a subscription while declining the retention offers and asking for the refund, watching a price until it drops, finding a hotel and booking once you say go, and signing into supplier portals each month to download invoices for an accountant.
Memory is the second thread. Nod reports 93.2 percent on LongMemEval, a public benchmark for long-term memory, among the better published results. It remembers months of conversation, reads how you write once Gmail is connected so drafts sound like you, and writes down how each task was completed so the next attempt starts from those notes and corrects them when a site changes. Everything it keeps can be viewed, edited or deleted.
Interaction stays in one conversation across surfaces: in the app, on Telegram or on WhatsApp, with several jobs running in parallel in the background and repeating ones such as a morning brief or a price check on a schedule. It also works ahead of you, reading connected accounts and offering jobs it could take off your plate.
The limits are in the release note. Running on a laptop, it stops when the machine sleeps, and the team suggests an always-on Mac mini or your own server while a hosted version stays on the roadmap. Each installation has one owner. It cannot make phone calls. One risk the team names itself: the permission check is also a model and can be wrong, which is why Nod asks whenever your words do not cover an action.
Key Features
- A standalone desktop app: Separate from n8n and usable without an n8n account or a server to set up first. You can connect an existing n8n instance if you want Nod to work with those workflows.
- Local sandbox with its own browser: Jobs run inside a private virtual machine on your computer with a dedicated browser. Data, memory and saved logins stay on the machine running it, and the model provider only receives what each task needs.
- Choose your own model: Point it at whichever provider you prefer with an API key or your ChatGPT plan, or connect an open-source model running on your own computer.
- A separate permission check: Before Nod sends, pays, posts or deletes anything you did not ask for, another model confirms you asked for it. A web page or an email cannot give it orders.
- Passwords stay out of the model: Saved logins and cards are encrypted on your machine and entered through private forms the model cannot read, so it sees the result rather than the credential.
- Long-term memory: 93.2 percent on LongMemEval, months of conversation retained, your writing style mirrored in drafts, and task notes that get reused and revised as sites change.
- One conversation, many surfaces: In the app, on Telegram or on WhatsApp, with several jobs running in parallel. Repeating jobs such as a morning brief run on a schedule, each showing an estimated monthly cost.
- Visible and capped cost: Every task shows what it cost and ends with a receipt, you can set a monthly spending limit, and Nod checks in once a task passes five dollars, a threshold you can change.
- Everything on record: Every sign-in, card use and action in your apps is logged alongside the cost of each task, so you can audit what it actually did.
Use Cases
- Cancel a subscription and hold the line on retention offers while getting the refund.
- Watch a product or flight price and report back at your target.
- Pick a hotel that fits the trip and book it after you confirm.
- Log into supplier portals monthly, download invoices and file them for an accountant.
- Pull numbers from a dashboard that has no export and put them in a weekly report.
- Hand off household errands such as renewals, appointments and following up on unanswered emails.
Pros
- Apache 2.0 with the rules in the code, so behaviour can be audited.
- Sandbox isolation plus credentials that never reach the model make the boundary legible.
- Cost is visible per task, with a monthly cap and a confirmation threshold.
- No n8n account and no server to stand up first, so the install is quick.
- Memory and task notes revise themselves, so recurring work gets cheaper over time.
Pricing
Nod itself is free and open source under Apache 2.0, with the code on GitHub. The cost is model usage, paid to whichever provider you connect, or nothing beyond electricity if you run an open-source model locally. The team notes that early testers found tasks cost more than expected and that recent releases brought it down substantially. Four guardrails keep spend bounded: every task ends with a receipt showing its cost, a monthly spending limit can be set, Nod asks before a task passes five dollars, and scheduled jobs display an estimated monthly cost.
Summary
Nod is built for the loose ends one person carries: cancelling a subscription that makes you argue with support, logging into supplier portals to pull invoices, comparing hotels before a trip. It is not an n8n client and needs no n8n account; you install it on your own machine and it brings its own local virtual machine and browser. Two things separate it from a chat assistant. Before acting it runs a separate model as a permission check, and a web page or an email cannot talk its way past that. And every task carries a cost, with a check-in once one passes five dollars. The trade-off is that it serves one person on one machine, and it stops when the laptop sleeps.
Version History
- n8n Labs releases Nod research preview (2026-11-10): n8n Labs, a new team inside n8n, released Nod, an open-source standalone desktop personal agent shipped as a research preview under Apache 2.0 for Mac and Windows, also deployable on your own server. It runs in a private local virtual machine with its own browser, works inside connected accounts such as Gmail, Calendar, Drive and Slack, and operates websites that offer no integration. The team reports 93.2 percent on LongMemEval for long-term memory, a separate model as a permission check before acting, credentials entered through private forms the model cannot read, a confirmation prompt once a task passes five dollars and an optional monthly spending limit. Known limits: tasks stop when a laptop sleeps, one owner per installation, and no phone calls.