Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

Nod

Open-source personal agent from the n8n team, with its own sandboxed computer and browser, that asks before it acts for you.

Reviewed by ReviewRamble editors · Updated

View on GitHub Open source

Share this page:XLinkedInFacebookRedditWhatsAppEmail

Our rating

3.9/5

Nod is the personal agent that treats your approval as a hard rule rather than a prompt, but it is a 0.8 research preview that sends app traffic through n8n's service by default.

What is Nod?

Nod is a personal agent from the team behind n8n, released as a research preview. It runs as a desktop app on macOS 14 or later, with Windows 11 and Linux x64 builds in preview, or as a self-hosted server. Each install is one Node.js server plus Docker containers: a Linux workspace that acts as a sandbox, a Chromium browser and PostgreSQL. It reaches Gmail, Google Calendar, Drive, Sheets, Docs, Slack and other apps through Composio, and Notion, Linear and n8n through their own MCP servers. You talk to it in the app, on Telegram or on WhatsApp.

The model is your choice: OpenAI, Anthropic, Gemini, OpenRouter, Bedrock, Vertex AI, Azure and 20+ other providers, a local server such as Ollama, or a ChatGPT Plus or Pro sign-in. A server-side trust check compares each app action, and each website step that can change something, against your messages and rules, and holds anything they do not cover for your approval. A Vault types passwords, codes and cards into websites without passing them to the model.

The repository, n8n-io/nod, was created on 25 September 2026. It is Apache-2.0 TypeScript and had about 2,440 GitHub stars on 18 November 2026. Releases come several times a week; v0.8.4 shipped on 17 November 2026.

Is Nod worth it?

Nod puts the safety of a personal agent at the centre of the design, and its documentation is unusually frank about where that safety runs out. OpenClaw runs tools on the host unless you configure a sandbox; Nod starts sandboxed. Its workspace runs as an unprivileged user with no access to your home folder, the Docker socket or provider keys, and its browser reaches only public addresses. Permission lives in the server, not the prompt. To run an action, the checking model has to quote the words of yours that allow it, and the server confirms those words exist before anything happens. Emails, pages and tool results reach the model marked as data and cannot grant permission.

n8n publishes its own evaluations. Nod finished 98% of 226 errands, run three times each, with no unsafe actions in those 678 runs; it scored 93.2% recall on the LongMemEval-S memory benchmark and formed no false memories from 216 WhisperBench attacks. Those runs used GPT-6.1 Sol in October 2026 on simulated accounts and websites, and the README calls them test results, not a guarantee. The choice of model matters more than usual: in 240 test trust decisions, GPT-6.1 Sol and Claude Opus 5.5 got every one right, while Claude Haiku 5.5 let 8 unsafe actions through. A cheaper checker saves money on the one part of the system that most needs to be right.

The security model page lists what the design does not cover. The checker is a model's judgement and is not immune to prompt injection. Once signed in to a website, Nod can act anywhere on it. The server and browser container can read secret values, and there is no hardware-backed key store. Learned skills are written from what Nod read, so a page or an email can shape one. By default, app calls pass through a connection service n8n hosts and through Composio, which holds your app access tokens; Composio runs with zero data retention for covered calls but still logs which tool ran and when. Usage data goes to n8n by default until you switch it off; conversation traces are off by default.

It is not yet an always-on assistant. Schedules run only while the machine running Nod is on, so continuous use means a Mac mini or a server; managed hosting is on the roadmap. The local runtime is about a 1.8 GB download on a Mac and 2.1 GB on Windows, which also needs WSL 2 and 20 GB of free disk. Nod sets a $200 monthly limit on API spending at setup and records the estimated cost of every model call. Against OpenClaw, Nod gives up reach, with far fewer chat channels and a much smaller community, for a tighter default posture. For anyone handing an agent a real Gmail account, that is the better trade.

Where it earns its place

  • A freelance consultant billing a dozen clients a month

    Chase unpaid invoices and walk into each client call prepared

    Connect Gmail and Google Calendar during setup; Nod drafts follow-ups and a one-page brief before each call, and sending waits for a yes unless a rule allows it. The software is free; an API model bills per task, capped by the default $200 monthly limit.

  • An operations lead at a ten-person agency

    Watch a supplier's price page and file month-end paperwork on sites that need a login

    Save the logins in Vault so passwords are typed into the page without reaching the model, set a schedule, and run Nod on a Mac mini so the schedule keeps going when laptops close.

  • A security engineer asked to approve Nod for staff

    Decide whether an agent may hold mail and website logins

    Read the limits section of the security model, require a capable checker model (Haiku 5.5 let 8 of 240 unsafe test actions through), switch off usage sharing, and decide whether app traffic may pass through n8n's connection service or one the team selects on a source install.

Before you commit

Setup
Download the macOS 14+, Windows 11 or Linux Flatpak build, connect a model and your apps; the README puts setup at about ten minutes. The local runtime is about 1.8 GB on a Mac and 2.1 GB on Windows, which also needs WSL 2 and 20 GB free. A source install needs Node.js 24.21 or later, pnpm and Docker.
Integrations
Gmail, Outlook, Google Calendar, Drive, Sheets, Docs, Slack and more through Composio; Notion, Linear and n8n over MCP; custom MCP servers, HTTP APIs and CLIs; chat in the app, Telegram or WhatsApp; 1Password for logins.
Data and privacy
Files, memory and saved logins stay on the machine running Nod; model providers receive what each task needs. App calls pass through n8n's hosted connection service and Composio, with zero data retention for covered calls. Usage data goes to n8n's PostHog project in the EU by default; conversation traces are opt-in, kept 14 days and never used for training.
Cost at scale
Nod is free; you pay your model provider. API spend is shown per call, with a $200 default monthly limit; a ChatGPT Plus or Pro plan avoids API billing. Always-on use needs a machine left running, such as a Mac mini or a server.
Limits
The trust check is a model and can be wrong; once signed in, Nod can act anywhere on that site; secrets are not hardware-protected; chat history cannot be cleared on its own; schedules pause when the host machine is off.

What’s good

  • Sandboxed workspace and browser by default; provider keys never enter either
  • Server-side trust check holds any action your words do not cover
  • Any model: 20+ providers, local servers or a ChatGPT plan
  • Published evaluations, including which models fail the safety check

What’s not

  • Research preview at v0.8, with several releases a week
  • App access goes through n8n's hosted service and Composio by default
  • Usage data is shared with n8n by default
  • Schedules stop when the computer running it is off

Best for

People who want an agent working in their email, calendar and browser and care more about approvals and sandboxing than about reach across chat apps.

Nod alternatives

Nod: common questions

Is Nod worth it?
Nod is the personal agent that treats your approval as a hard rule rather than a prompt, but it is a 0.8 research preview that sends app traffic through n8n's service by default. We rate it 3.9 out of 5.
How much does Nod cost?
Free and open source (Apache-2.0). You pay your model provider directly, or sign in with a ChatGPT Plus or Pro plan.
What are the downsides of Nod?
Research preview at v0.8, with several releases a week; app access goes through n8n's hosted service and Composio by default; usage data is shared with n8n by default; schedules stop when the computer running it is off.
Who is Nod best for?
People who want an agent working in their email, calendar and browser and care more about approvals and sandboxing than about reach across chat apps.
How much does Nod cost for a team or in production?
Nod is free; you pay your model provider. API spend is shown per call, with a $200 default monthly limit; a ChatGPT Plus or Pro plan avoids API billing. Always-on use needs a machine left running, such as a Mac mini or a server.
Is Nod safe for company data?
Files, memory and saved logins stay on the machine running Nod; model providers receive what each task needs. App calls pass through n8n's hosted connection service and Composio, with zero data retention for covered calls. Usage data goes to n8n's PostHog project in the EU by default; conversation traces are opt-in, kept 14 days and never used for training.
What are the best Nod alternatives?
The closest alternatives we have reviewed are OpenClaw, Hermes Agent, nanobot and Eigent.

Answers checked Nov 18, 2026.