Computers and Software

Personal AI Agent, Setting Up Nod by n8n on a Mac

How to Set Up Nod: Free Personal AI Agent by n8n (Mac)

Video: How to Set Up Nod: Free Personal AI Agent by n8n (Mac)

ยท based on the channel Mike Murphy | AI Handyman

Key takeaways

  • Nod is a free, open-source personal AI agent published by n8n on GitHub at github.com/n8n-io/nod.
  • The Mac install is a download from GitHub followed by dragging the app into the Applications folder.
  • Gmail is connected through Composio, the integration layer Nod uses for third-party apps.
  • Nod can run on an existing ChatGPT subscription, so no separate API key or per-token billing is required.
  • Six sidebar sections cover Chat, Scheduled, Apps, Channels, Usage and Vault, reachable with Cmd+1 through Cmd+6.

Nod is a free, open-source personal AI agent from n8n that you install as a desktop app on macOS, connect to your own accounts, and drive from a chat window or a schedule. Setup takes a few minutes: download the build from GitHub, drag it into Applications, run the onboarding wizard, link Gmail through Composio, and authenticate with your existing ChatGPT subscription instead of buying API credits. The full walkthrough, with screenshots and the same steps in order, lives on the Nod tutorial page.

What Nod actually is and why it runs locally

Nod is an agent, not a chatbot wrapper: it holds credentials, calls tools, and executes tasks on a schedule without you sitting in front of it. Because it is distributed by n8n โ€” the company behind the widely used workflow automation platform โ€” the design assumes you will bolt on connectors rather than accept a fixed feature list. The app is open source, so the code is inspectable and the integration surface is not gated behind a paid tier.

Running it on your own Mac matters for two reasons. First, tokens and credentials stay on your machine rather than passing through a vendor's server. Second, the agent can reach local resources โ€” files, browser sessions, command-line tools โ€” that a hosted assistant cannot touch. The trade-off is that the Mac has to be awake for scheduled jobs to fire; a laptop that sleeps at night will miss a 7 a.m. task.

Install Nod on a Mac in three steps

Nod is not in the Mac App Store, so installation is a manual drag-and-drop from the GitHub release page.

  1. Open https://github.com/n8n-io/nod and download the macOS build from the releases section.
  2. Open the downloaded disk image or archive and drag the Nod icon into the Applications folder.
  3. Launch Nod from Applications. On first start macOS may ask you to confirm opening an app from an identified developer; approve it and the setup wizard appears.

The wizard is interactive: it asks questions in a chat-style interface and fills in configuration as you answer. If you break something later or want to add an account you skipped, the onboarding wizard can be replayed from settings without reinstalling.

Connect an AI provider and pick a model

Nod needs a language model before it can do anything useful, and the cheapest route is the subscription you already pay for. During setup you choose an AI provider and then a specific model; selecting "ChatGPT" lets Nod authenticate against an existing ChatGPT plan rather than requiring a separate OpenAI API key with per-token billing. That distinction is the one people underestimate โ€” the provider choice determines both cost and capability, and the model choice inside it determines how well the agent handles multi-step tool calls.

A practical sequence:

  • Pick the provider that matches a plan you already hold, so testing costs nothing extra.
  • Choose a mid-tier model for routine scheduled jobs and a stronger one for research tasks that chain several tools.
  • Keep an OpenAI-compatible endpoint in mind as a fallback if you later want to route requests through a different gateway or a compute layer you already use.
  • Revisit the model setting after a week of usage; the Usage panel shows what each job actually consumes.

A tour of the six sidebar panels

Every core function sits behind a keyboard shortcut, which makes Nod fast to operate once you learn six combinations.

  • Chat (Cmd+1) โ€” the main conversation window where you talk to the agent and it calls tools mid-answer.
  • Scheduled (Cmd+2) โ€” recurring jobs, each with a prompt, a cadence and a delivery target. This is where a weekly AI news roundup lives.
  • Apps (Cmd+3) โ€” connected services. Gmail arrives here once Composio finishes linking your inbox.
  • Channels (Cmd+4) โ€” messaging surfaces such as Telegram and WhatsApp, so you can message the agent from your phone instead of the desktop.
  • Usage (Cmd+5) โ€” token and request accounting per task, useful for spotting a schedule that burns more than it returns.
  • Vault (Cmd+6) โ€” stored credentials and secrets, kept separate from the chat history.

Connecting Gmail, custom MCP servers and messaging channels

Gmail is the first integration most people add, and it goes through Composio, the connector layer Nod uses for OAuth-based apps. You authorise the inbox once and the agent gains read and search access, which is what makes inbox-summary and follow-up tasks possible.

For anything not in the Composio catalogue, Nod accepts custom MCP servers โ€” the Model Context Protocol is the standard interface the agent uses to discover tools. Adding one means pointing Nod at a server endpoint and letting it enumerate the available actions; from then on those actions appear to the model exactly like built-in ones. This is the escape hatch that keeps the agent from being limited to a fixed app list.

Channels extend reach beyond the desktop. Linking Telegram or WhatsApp means a scheduled job can push its result to your phone, and a quick question can be answered from a queue or a train without opening the Mac app. Note that the Mac still has to be running to service the request.

Building a first scheduled task

A scheduled task in Nod is a prompt plus a trigger, and the canonical first example is a weekly AI news roundup. In the Scheduled panel you create a new job, write the instruction โ€” for instance, search for notable AI tool releases from the past seven days, deduplicate against last week's list, and summarise in five bullets โ€” set the cadence to weekly, and choose where the output goes: the Chat panel, an email, or a messaging channel.

Two habits make scheduled jobs reliable. First, be explicit about the output format in the prompt, because the agent will otherwise vary its structure week to week. Second, check the Usage panel after the first two runs; a research job that fans out across many searches can consume far more tokens than a summarisation job, and the model choice is the lever you adjust. Existing tasks can be edited in place, so tuning a prompt does not mean rebuilding the schedule.

Common questions and pitfalls from early users

Provider and model selection is the step that trips people up most, because the two settings are separate and the wrong pairing either costs money or produces an agent that fumbles tool calls. Choose the provider first, confirm the authentication succeeded, then pick the model.

Other recurring snags:

  • Sleeping laptops. Scheduled jobs do not run while the Mac is asleep. If a task must fire reliably, either keep the machine awake or run the agent on an always-on machine such as a small VPS.
  • OAuth expiry. Gmail access through Composio can lapse if you change your Google password; re-authorise from the Apps panel.
  • Token burn during testing. Iterating on a prompt with a top-tier model gets expensive fast on API billing; using a subscription plan or an OpenAI-compatible gateway keeps experimentation cheap.
  • Over-broad permissions. Grant the narrowest scopes that make the task work, especially for inbox access.

Summary

Nod gives you a personal AI agent that runs on your own Mac, connects to real accounts through Composio and MCP, and reuses a ChatGPT subscription so the model layer costs nothing extra. The install is a GitHub download and a drag into Applications; the rest is the onboarding wizard, a Gmail authorisation, and a first scheduled job. Six sidebar panels โ€” Chat, Scheduled, Apps, Channels, Usage and Vault โ€” cover everything from conversation to credential storage. The step-by-step video breakdown was produced by the channel Mike Murphy | AI Handyman, and the written companion with the same sequence is available at the Nod tutorial page.

Source: How to Set Up Nod: Free Personal AI Agent by n8n (Mac) ยท Markdown version

Questions & answers

Do I need an OpenAI API key to use Nod?

No. Nod can authenticate against an existing ChatGPT subscription, so the model layer adds no per-token cost during setup and testing. An API key or an OpenAI-compatible endpoint is only useful if you want to route requests through a different provider or gateway.

Will scheduled tasks run while my Mac is asleep?

No. Nod runs locally, so a sleeping or powered-off Mac cannot fire a scheduled job. For tasks that must run reliably at a fixed time, keep the machine awake or host the agent on an always-on server such as a small VPS.

How do I connect an app that is not in the integration list?

Use a custom MCP server. Nod speaks the Model Context Protocol, so you point it at a server endpoint, it enumerates the available tools, and those actions become callable by the agent just like built-in integrations.

Can I talk to Nod from my phone?

Yes, through the Channels panel. Linking Telegram or WhatsApp lets you message the agent and receive scheduled output on your phone, though the Mac running Nod still has to be awake to answer.