BetaEditorial Listing

Nod

n8n · Nod: Open-Source Personal Agent From n8n Labs With Its Own Sandboxed Computer, Browser and Memory

Open Nod

Nod is a free, open-source (Apache 2.0) personal agent from n8n Labs that runs on your Mac, Windows PC or your own server, with its own sandboxed computer and browser, to do errands such as chasing invoices, preparing call briefs, watching web pages and cancelling subscriptions. You bring the model, and a separate permission check makes it ask before actions your request does not cover. It suits freelancers, solo operators and technical individuals comfortable connecting an AI provider key.

PricingFree
Setupmedium
Runs onDesktop · Self-hosted
Open sourceYes
DocsYes
CategoryProductivity
Personal AssistantComputer UseBrowser AutomationPersistent MemoryHuman-in-the-LoopOpen SourceSelf-HostedLocal-FirstDesktop AppSandboxModel-AgnosticBYOKLocal ModelsOllamaScheduled TasksTelegramWhatsAppMCPEmail

Best for

Freelancers, solo operators and technical individuals who want an agent to take recurring errands across their own apps and websites, keep it on hardware they control, choose and pay for their own model, and approve anything consequential

Not ideal for

Teams that need one shared assistant, people who want an always-on agent without keeping a computer or server running, anyone unwilling to manage an AI provider key or ChatGPT plan, Linux desktop users who do not want to run a server, and organizations that need a supported, production-grade product rather than a research preview

Who it's for

Freelancers, solo business owners and technically comfortable individuals who handle their own client work, follow-ups, invoices and life admin

Capabilities

  • Its own computer: a locked-down Linux workspace container and a Chromium browser container, run inside a private virtual machine in the desktop app or as Docker containers on a source or server install
  • Works on real websites with no integration: signs in with saved logins, fills forms, downloads files and completes checkouts, and can hand a browser session to you when it needs you
  • Separate trust check that reviews each app action and each browser step that can change something against your messages and rules; reads and drafts run, and anything your words do not cover waits for your yes
  • Every payment waits for your approval with the card's security code, which Nod does not save
  • Vault that stores logins, verification codes and cards encrypted on your machine and enters them into the site privately, so the model sees account names but never the secrets; a 1Password helper can supply logins
  • Content from apps, web pages and emails reaches the model marked as data, not instructions, and the trust check sees your words and the action rather than the pages Nod read
  • Long-term memory with every note visible and editable alongside the message it came from; learns only from your own messages, not from pages or emails it reads
  • Learns your writing style from sent email for drafts, and saves the steps it used for a task so it can reuse them and fix them when a site changes
  • One conversation in the desktop app, Telegram or WhatsApp, with several independent tasks running in parallel in the background
  • Recurring schedules such as a morning brief or a price check every few hours, with an estimated monthly cost shown for each schedule
  • Proactive mode that reads your connected apps and offers jobs it could take on, without acting until you agree
  • App connections for Gmail, Outlook, Google Calendar, Drive, Sheets, Docs, Slack and many more through Composio, plus Notion, Linear and n8n through their own MCP servers and any MCP server address you add
  • Bring your own model from more than 30 providers, including OpenAI (API key or ChatGPT Plus or Pro sign-in), Anthropic, Google Gemini, OpenRouter, Amazon Bedrock, Vertex AI and Azure OpenAI, or a local model through Ollama, LM Studio, vLLM or llama.cpp
  • Usage view that logs every model request, app action, sign-in and card use; with API models it adds estimated costs, a per-task receipt, a monthly spending limit and a check-in when a task passes $5 (adjustable)

Limitations

  • Research preview from an experimental n8n Labs team; n8n says the permission check is itself a model's judgement and can make mistakes
  • Safety depends on the model you choose: in n8n's own test of 240 permission decisions, GPT-6.1 Sol and Claude Opus 5.5 got all of them right while Claude Haiku 5.5 let 8 unsafe actions through, and n8n says small local models are less reliable at the check
  • Stops when the computer running it sleeps or shuts down; always-on schedules need a Mac mini or your own server, and a managed hosted version is only on the roadmap
  • Each installation has one owner, with no multi-user or team setup
  • Cannot make phone calls; calling businesses is listed as a future feature
  • Windows 11 support is a preview and a local assistant needs WSL 2 with hardware virtualization and at least 20 GB of free disk; there is no Linux desktop installer, so Linux means a source or self-hosted server install
  • You pay your model provider directly, and n8n notes that some early testers found tasks cost more than expected
  • Local models need tool calling and at least 64,000 tokens of context, and the trust check still uses a model (your local one by default, or a small model or decision service from another provider)
  • Claude needs an Anthropic API key; Claude and Gemini app subscriptions cannot be used to sign in, unlike ChatGPT Plus or Pro
  • App connections pass through an n8n-hosted connection service and Composio by default, and basic usage data is shared with n8n by default until you turn it off in Settings
  • WhatsApp chat uses Baileys, an unofficial WhatsApp client, which the docs say carries a small risk of account restrictions

Use cases

  • Chasing an unpaid invoice with friendly reminders, with Nod asking before a second nudge
  • Getting a one-page brief each morning for every call, built from email threads, the calendar and the client's latest news
  • Watching a competitor's or supplier's pricing page every few hours and hearing only when it changes
  • Cancelling a subscription, declining the retention offers and requesting a refund
  • Signing in to supplier portals each month to download invoices and file them for an accountant
  • Turning client requests from email and Slack into Linear tasks while skipping duplicates

Our take

OpenClaw and Hermes Agent give technical users a lot of freedom. Nod narrows the job to errands and builds the guardrails into the server rather than the prompt: a sandboxed workspace and browser, a vault the model cannot read, a permission check that ignores page and email content, approval on every payment, and a log with costs for everything it did. That makes it a credible local alternative to hosted errand agents for people who do not want their logins and memory in someone else's cloud. The catch is that the guardrails are only as good as the model behind them, and n8n's own numbers show a smaller model letting unsafe actions through. It is also a research preview that stops when your laptop sleeps, serves one person per install, and bills every step to your own model provider or ChatGPT plan unless you run a local model. Try it on a Mac or a spare always-on machine with a frontier model and a monthly limit set, and hand it one recurring errand before trusting it with your inbox.

Who should use it

Freelancers and solo operators with recurring admin across email, portals and dashboards; privacy-minded individuals who want an errand agent whose data and credentials stay on their own hardware; and n8n users or developers who want an open-source agent they can inspect and change.

Who should skip it

Teams that need a shared assistant, people who want a zero-setup consumer product or an agent that keeps working with no machine left on, Windows users who cannot enable WSL 2, and anyone who needs phone calls or production-level support.

Strengths

  • Free and open source under Apache 2.0, with the permission rules in the published code
  • Data, memory, files and saved logins stay on the machine or server that runs Nod
  • Has its own browser, so it can finish errands on websites that offer no API or integration
  • Permission checks and the credential vault run in the server rather than in the prompt, and payments always need your approval
  • Wide model choice, including ChatGPT Plus or Pro sign-in and local models through Ollama, LM Studio, vLLM or llama.cpp
  • Visible, editable memory and a usage log with per-task costs and a monthly spending limit

Weaknesses

  • Research preview, and the permission check is a model's judgement that is weaker with small models
  • Stops when the host computer sleeps unless you keep a Mac mini or server running
  • One owner per installation and no phone calls
  • Windows is a preview that needs WSL 2, and there is no Linux desktop app
  • Model costs are yours, and n8n says some early testers found tasks cost more than they expected
  • App connections go through an n8n-hosted service and Composio by default, and usage data sharing is on by default

Nod pricing

Nod (open source)

$0

  • Free download for macOS 14+ and Windows 11 (preview), or self-host from source
  • Apache 2.0 license
  • Bring your own AI provider key, a ChatGPT Plus or Pro plan, or a local model
  • Per-task costs, receipts and a monthly spending limit on API models

Note: Nod itself does not charge you. Your model provider bills you for the requests Nod makes with your key or plan; a local model costs nothing per request. Costs shown in Usage are Nod's estimates from each model's API price, and by default Nod checks in once a task passes $5. A managed hosted version is on the roadmap, with no published pricing.

Technical specs

Available models

GPT-6.1 Sol (OpenAI API or ChatGPT Plus/Pro sign-in)Claude Opus 5.5 (Anthropic API)Gemini 3.8 Flash (Google AI Studio)Gemini 4 Argon (Vertex AI)Amazon Bedrock and Azure OpenAI models30+ providers including OpenRouter, Groq, Mistral, DeepSeek and xAILocal models via Ollama, LM Studio, vLLM or llama.cpp (tool calling and 64K+ context required)

Where Nod excels

Month-end paperwork for a freelancer

Nod can sign in to bank, tax and supplier portals with saved logins it never shows the model, download the statements and invoices, and file them by name, which a chat assistant would hand back as a to-do list.

Client follow-up without dropped questions

It reads email and Slack, finds a question that never got an answer, drafts the reply from your own documents, and holds the send for your approval when your instructions do not already cover it.

Price and page monitoring

A schedule checks a page every few hours and speaks up only on a change, with the schedule's estimated monthly model cost shown up front.

Life admin such as refunds and bookings

With its own browser it can work through cancellation flows and booking sites that have no integration, and every payment waits for your approval.

Nod vs. competitors

Nod vs. OpenClaw

OpenClaw is a free, open-source personal assistant that lives in 20+ messaging apps and suits technical users who want broad control over what the agent can do. Nod is a packaged desktop app focused on errands, with its own sandboxed workspace and browser, a credential vault the model cannot read, and a server-side permission check, but you reach it only through its own desktop or web app, Telegram and WhatsApp.

Nod vs. Hermes Agent

Hermes Agent is Nous Research's open-source, self-hosted agent that builds memory and reusable skills across sessions, aimed at developers. Both are model-agnostic and keep memory on your machine; Nod adds a consumer desktop installer, an isolated browser for website errands, and approval and spending controls, while Hermes gives developers a more general-purpose agent to build on.

Nod vs. Tab

Tab is a personal errand agent you text over iMessage or WhatsApp; it runs on its own cloud computer, can place phone calls and pays only after you approve the exact total, and is in private beta. Nod is free and open source, runs on your own computer or server with your own model, and cannot make phone calls yet.

Nod vs. OpenAI Dots

OpenAI's dots are always-on agents inside ChatGPT Pro, Business Premium and Enterprise, each with its own cloud computer and browser, that keep working on projects between conversations. Nod runs on hardware you control with a model you choose, including a ChatGPT Plus or Pro plan, but stops when that machine sleeps.

Nod vs. Muse

Meta's Muse is a consumer personal agent that does errands in a dedicated cloud VM and asks before sensitive actions. Nod does similar errands but keeps its sandbox, memory and saved logins on your own machine, is open source, and requires you to bring an AI provider.

Frequently asked questions

What is Nod by n8n?

Nod is an open-source personal agent from n8n Labs, an experimental team at n8n. It runs on your computer or your own server with its own sandboxed workspace and browser, works in your connected apps and on ordinary websites, remembers what you tell it, and asks before actions your request does not cover. n8n released it as a research preview on November 10, 2026.

Do I need n8n to use Nod?

No. Nod is a separate app and does not require an n8n account. It is not the n8n AI Assistant or an agent built inside n8n. If you already run n8n, you can connect your instance so Nod can work with your workflows.

Is Nod free and open source?

Yes. The code is on GitHub under the Apache 2.0 license and the app is free to download. You pay your AI provider for the model requests Nod makes, or use a ChatGPT Plus or Pro plan, or run a local model at no per-request cost.

Which models does Nod support?

More than 30 providers, including OpenAI (API key or ChatGPT Plus or Pro sign-in), Anthropic (API key), Google Gemini, OpenRouter, Amazon Bedrock, Google Vertex AI, Azure OpenAI and Cloudflare, plus any OpenAI-compatible server. Local models work through Ollama, LM Studio, vLLM or llama.cpp if they support tool calling and at least 64,000 tokens of context.

Which computers does Nod run on?

The desktop app runs on macOS 14 or later (Apple silicon or Intel) and, in preview, on Windows 11 x64 with WSL 2. You can also build it from source and host it with Docker on a Linux AMD64 or ARM64 server or a Mac, for one person, and connect to it from the desktop app or a browser. There is no Linux desktop installer.

Does Nod ask before it acts?

It asks whenever your words do not cover an action. Reads and drafts run directly. Sending, posting, deleting and other changes go through a separate check that compares the action with your request and rules and asks you if they do not match, and every payment needs your approval with the card's security code. The check is a model's judgement, so n8n recommends a capable model.

Can Nod run when my laptop is closed?

No. Nod and its schedules run only while the computer hosting it is on and awake. To keep it running, n8n suggests an always-on Mac mini or your own server. A managed hosted version is on the roadmap.

How well does Nod perform?

n8n reports that, with GPT-6.1 Sol in October 2026, Nod finished 98% of 226 errands run three times each in simulated accounts and websites, with no unsafe action in those 678 runs, and scored 93.2% on the LongMemEval-S memory benchmark. These are vendor test results in simulated environments, and n8n notes that results depend on your model and the sites you use.

Integrations & fit

GmailOutlookGoogle CalendarGoogle DriveGoogle SheetsGoogle DocsSlackNotionLinearn8nTelegramWhatsApp1PasswordComposioMCPOpenAIChatGPTAnthropicGoogle GeminiOpenRouterAmazon BedrockGoogle Vertex AIAzure OpenAIOllamaLM StudiovLLMllama.cpp
Good fit forSolo / individual, Startup / small team
Pricing modelFree· No cost to start
See pricing on Nod →

Alternatives to consider

About Nod

Nod is a standalone app from n8n Labs, an experimental team inside n8n, released as a research preview on November 10, 2026. It is separate from the n8n automation platform and needs no n8n account: n8n positions it as a single-player tool for one person's jobs rather than the shared team processes n8n runs. You install the desktop app, connect a model and your apps, and talk to Nod in one conversation in the app, on Telegram or on WhatsApp while it runs several tasks in parallel in the background. Because it has its own Linux workspace and Chromium browser, it can finish work on websites that have no integration, such as signing in to supplier portals to download invoices, pulling numbers from a dashboard with no export, or cancelling a subscription and turning down the retention offers. It also works through connected apps (Gmail, Outlook, Google Calendar, Drive, Slack, Notion, Linear and others via Composio and MCP), runs recurring jobs on a schedule, and reads your connected apps to suggest work it could take on. Memory is a core feature: it keeps notes from months of conversation that you can see, edit or delete, learns your writing style from sent email, and saves how it did a task so it can reuse and repair the steps when a site changes. The trust design is what most separates it from hosted errand agents. Everything runs on the machine that hosts Nod (a private virtual machine in the desktop app, Docker containers for source and server installs), saved logins and cards are encrypted locally and entered into sites by the server so the model never sees them, and a separate model-based check reviews each app action and state-changing browser step against your own words and rules, never against page or email content, and asks you when they do not cover it. Every payment needs your approval with the card's security code. The main tradeoffs: you pay your own model provider, the check is only as reliable as the model behind it, Nod stops when the computer running it sleeps, and each installation has one owner.

Updates from Nod

LaunchNod released as a research preview

n8n Labs released Nod, a standalone open-source personal agent with its own sandboxed computer and browser, as a research preview for macOS and Windows (preview), with self-hosting from source. The same-day 0.6.1 release cut the cost of background checks and showed background spend separately in Usage.

Are you the founder? Claim this listing →