n8n Labs has put out a research preview of Nod, a standalone personal agent you run on your own computer. The app is open source, talks to the connected services you allow, and can work through an ordinary website in its own browser. You pick the AI provider. You don’t need an n8n account, and Nod isn’t the n8n product itself.
Nod v0.6.2 is the build on the download page now, with Mac and Windows installers. The code is on GitHub under the Apache 2.0 licence.
Meet Nod, your own open-source personal agent from n8n Labs
— n8n.io (@n8n_io) November 11, 2026
Ask it to prep your calls, collect invoices or cancel a subscription
It runs on your computer with its own browser to get the job done across apps and websites, using the AI provider you choose👇 https://t.co/muB2JGFywG
From n8n Labs
On 10 November, Ophir Prusak wrote on the n8n blog that Nod comes from n8n Labs, "a new team within n8n that experiments with ideas outside our core product." He says the experiment leaves n8n’s direction alone, and that work on n8n and the n8n agent is continuing as planned.
Johan Cutych, writing from Labs on the community forum, draws a sharper line of use. n8n is for the shared processes of a team or a business. Nod is for one person, doing that person’s own jobs. A connected n8n instance is optional. If you hook one up, Nod can build and test workflows there, but a new workflow stays a draft until you publish it.
Jobs it can run
Because the agent has a local sandbox and a browser, n8n is pitching work that usually stalls in a chat window. Cancel a subscription, refuse the offers to stay, and ask for a refund. Watch a price and speak up when it moves. Find a hotel, then book it only after you agree. Log in to a supplier portal each month for the invoices. Pull figures off a dashboard that won’t export.
You can talk to the same conversation from the app, the web, Telegram or WhatsApp, and more than one job can run in the background. A morning brief or a price check can be scheduled. Nod can also look through connected apps and suggest work. The docs name Gmail, Calendar, Slack, Notion and Linear among the default connections, and they say you can take over the browser when a site needs you.
n8n Labs ships Nod as a research preview: an open-source personal agent that runs locally with its own computer and browser.
— Documenting AGI (@DocumentingAGI) November 11, 2026
It finishes multi-step jobs (call prep, invoice collection, subscription cancellations) across apps and websites, asks before acting outside your rules, https://t.co/Ngd9YRAwCQ
Where it runs, and when it stops
macOS 14 or later is supported, on Apple silicon or Intel. Windows 11 on x64 is a preview, and a local Windows assistant also needs WSL 2, hardware virtualization, and at least 20 GB of free disk. The install documentation puts the app and runtime at about 2.1 GB to download. n8n GmbH signs the installer. Microsoft’s SmartScreen can still warn, because the release is new.
Linux users don’t get a desktop app yet. A source install works on macOS and Linux, or you can stand up a server with Docker Compose and reach it from the web or the desktop app. Several devices can talk to one server. Each installation still has a single owner, and two separate installs don’t share their data.
Local work stops if the machine sleeps. For anything that has to keep to a timetable, the docs and the blog point to an always-on Mac mini or a self-hosted server. Managed hosting is on the roadmap, as are phone calls and a way for a few people to share one Nod. Phone calls don’t work yet.
It asks first, and that check can be wrong
Nod’s Trust layer reviews app actions and website steps that can change something, matches them against your request and rules, and asks when those words don’t cover the action. Every payment waits for your approval. A web page or an email can’t grant permission. Only you can, through your messages, approvals and standing instructions.
The permission check is itself a model and can make mistakes, which is why Nod asks when your words don’t cover an action.
Ophir Prusak, n8n blog
Background work that Nod starts by itself only reads and drafts, unless you’ve already told it what to do on its own. Saved logins, authenticator keys and cards sit in Vault. The server types those values into the site through private forms, so the model sees account names and status, not the secret. On a local desktop app, the Vault key is held in the system’s safe storage.
What it costs
Nod is free. The bill is the model. You can use OpenAI, Anthropic, Gemini, OpenRouter, Amazon Bedrock, Vertex AI, Azure, a ChatGPT Plus or Pro login, or a local OpenAI-compatible server such as Ollama.
If you use an API key, Usage estimates the spend and you can cap the month from US$1 upwards. Hit the cap and new chat, schedules and background jobs stop before the next model call. Work already in flight can finish, and an action you’ve already approved still runs once. There’s a separate check-in, set at US$5 by default, when a single task’s spend since your last message crosses that line. You can change the amount. ChatGPT plan use is priced at API rates for those check-ins, even though it doesn’t raise a second bill.
The blog mentions early testers who were surprised by the cost, and says a recent release cut it substantially. It doesn’t name the testers or give the figures.
The scores n8n published
On the company’s own tests, with GPT-6.1 Sol, Nod finished 98% of 226 errands in simulated apps and websites. Each errand had three attempts, 678 runs in all, recorded on 17 October, and n8n says none of those runs had an observed unsafe action. Two days later, on LongMemEval-S, it scored 93.2% on 500 questions asked in fresh sessions. On WhisperBench it recorded no memory injections across 216 runs: 108 attack emails, tried once as a direct check and once with background monitoring on.
These are n8n’s measurements, on one model and a named source revision, in a simulated world. The evaluation report says the 98% is rounded and measures task completion, not a pass on every honesty, safety or product check. Thirteen prompt-injection errands sit inside the 226, not as a separate survey of attacks. In 12 WhisperBench later answers, the planted claim turned up because Nod read the attack email again from the inbox, not because it had stored the claim as memory. The same report says a current run does not recreate the October task snapshot, and that zero observed unsafe actions is not a guarantee of zero risk in production.
We’ve just shipped Nod, your own open-source personal agent from n8n Labs.
— Johan Cutych (@johancutych) November 11, 2026
How is this different from everything else out there??? 🧵 https://t.co/WbwKeRL0a7
What else to know
This is an early preview, so features and data formats can still change. Files, memory and saved logins stay on the computer or server that runs Nod. Model services, apps and websites get the data they need for each task. App connections use n8n and Composio by default. Nod also checks for updates and shares usage data with n8n, unless you turn that off in Settings.
You can download Nod at nod.n8n.io, read the docs, or inspect the code on GitHub. Testers are asked to use Send feedback in the app, or to post in the n8n Labs category on the community forum.
Discover more from Sulat AI & Tech News
Subscribe to get the latest posts sent to your email.
