N8n's New Agent Nod Runs on Your Laptop, Free and Open Source

N8n shipped Nod on November 10, 2026, a free, open-source personal agent for Mac and Windows that runs tasks in a user's apps and on websites from a private sandbox on that person's own computer. The release came from n8n Labs, a new team for experiments outside the core product, and lands as a research preview.

The code ships under Apache 2.0. Nod needs no n8n account, server or cloud-service setup — though users must connect an AI provider. N8n said development of its core product and the n8n agent continues as planned.

What can Nod actually do?

Nod has its own browser and can sign in to websites that have no formal integration. It is not an n8n workflow and not the n8n AI Assistant, though existing users can optionally connect an instance so the agent works with their workflows. The product broadens n8n's automation focus from shared team processes to jobs handled by one person — business or household.

Concrete examples from the documentation and announcement:

• Cancelling a subscription and requesting a refund
• Retrieving invoices from supplier portals
• Watching prices and extracting figures from dashboards without export functions
• Web and PDF research
• Creating documents, spreadsheets and code in a Linux workspace
• Reading or drafting in Gmail, Calendar, Slack, Notion and Linear

Users keep one conversation across the app, Telegram or WhatsApp while several jobs run in parallel. Recurring work — morning briefs, price checks — can run on schedules, and Nod can propose jobs after reading connected apps.

How good is its memory?

Here's the number that stands out: n8n's GitHub repository reports a 93.2% LongMemEval score, up from 79%, using 500 questions about earlier chats answered in a fresh session. Nod retains months of conversation, records how it completed tasks, and reuses and revises those notes when a site changes. It can also learn a user's writing style from sent Gmail messages, and users can inspect, edit or delete stored information.

The LongMemEval methodology tests information extraction, multi-session reasoning, knowledge updates, temporal reasoning and abstention when information is unknown. Translation: the score concerns long-term memory, not how reliably Nod finishes a booking or refund.

Who pays, and what's the catch?

Before Nod sends, pays, posts or deletes anything beyond the user's request, a separate model-based check asks for confirmation. A web page or email cannot grant it authority. N8n cautions the check is itself a model and can make mistakes — a safeguard, not a guarantee.

On the data side: saved logins and payment cards are encrypted on the machine and entered through private forms the model cannot read. The database, files, memory and logins stay local; the selected AI provider receives only what each task needs. Beyond update checks, information reaches n8n only when usage data is enabled or feedback is submitted. Nod also logs sign-ins, card use, connected-app actions and model costs.

Nod itself is free, but users pay their chosen model provider — or connect an open-source model running locally. Supported connections include OpenAI, Anthropic, OpenRouter, Amazon Bedrock, Google Vertex AI, Azure OpenAI, plus ChatGPT Plus and Pro plans. Cost controls include a receipt per finished task, a monthly spending limit, an adjustable check-in once a task passes $5, and an estimated monthly cost for each schedule.

Where the preview falls short

The desktop download requires macOS 14 or later; Windows 11 support is labeled a preview, and self-hosting on a personal server is available. A laptop installation stops when the laptop sleeps, so scheduled jobs need the machine on. Each installation has one owner, and Nod cannot make phone calls. N8n lists a hosted version and shared team use as future work.

The bet is clear: local control and provider choice, traded against direct model costs, an always-on machine for unattended schedules, and an approval mechanism that can still err.