What we cover:
- What Nod is, and why it's not another chatbot
- How Nod handles trust, and what its test numbers actually say
- What my article pipeline looks like today
- The plan to move it into Nod, in three steps
- The two problems to solve before it's real
Most AI tools hand you a plan and wait for you to do the work. Nod might actually finish the job. That's the claim, anyway, and it's why I'm writing about it before I've even installed it.
n8n Labs released Nod on November 10, 2026. It's a standalone personal agent that runs on your own computer, with its own computer and browser inside it. You tell it what you need in plain language, and it works in your apps and on real websites until the job's done. It's open source under Apache 2.0, it needs no n8n account, and you bring your own model.
What is Nod?
A chat assistant hands you a plan and waits. A workflow tool like n8n does exactly what you built, nothing more. Nod sits between them. It takes a job, figures out the steps, and finishes it, including the parts that happen in a browser with a login and a form.
The examples n8n gives are everyday errands:
- Call briefs built from your threads, your calendar, and the client's latest news
- Chasing an invoice, with a friendly reminder and a pause before the second nudge
- Cancelling a trial, turning down the retention offers, and asking for a refund
- Watching a page and telling you only when something changes
Why trust is the whole question
This is the part I care about, because an agent that can act on websites is an agent that can do damage. Nod asks before it acts. A separate check reviews each app action and each browser step that could change something against your own words. Reads and drafts run free. Anything your words don't cover waits for your yes.
Logins, verification codes, and cards live in a vault on your machine, encrypted. The model sees account names but never the secrets. Every payment waits for approval with the card security code, which Nod doesn't save. Data, memory, and saved logins stay on the machine running it.
The numbers n8n publishes: 98 percent of 226 test errands finished across three runs each, zero unsafe actions in 678 runs, and 93.2 percent on LongMemEval, a public test of long-term memory. Those are their figures from simulated accounts, not independent verification, and I haven't run Nod myself yet. So treat them as the design goal, not a guarantee.
What does it cost?
Nod itself is free. You pay your model provider, or you connect a ChatGPT Plus or Pro plan, or you run a local model through Ollama or LM Studio. It shows a receipt for each finished task, a monthly spending limit, and it checks in once a task passes $5. Each schedule shows an estimated monthly cost before you turn it on.
What my article pipeline looks like today
I keep a list of topics in a Google Sheet, one row per idea, with a tone guide that says how I write. A daily run turns the next row into a draft on austincodez.dev, my Drupal site, through the JSON:API with a Key Auth key. Each draft gets an AI image in a consistent house style, tags, and internal links. It stays a draft until I read it and publish it myself. A Sunday session adds ten new ideas to the sheet.
That machine works. But the daily run still needs me or a helper in the middle of it.
I also looked at ChatGPT scheduled tasks for the same job. They can draft on a schedule, but they can't post to Drupal or edit the sheet on their own, so the last mile is still manual. That's the gap Nod might close.
The migration plan
The plan is to move the daily run into Nod, in pieces, and keep the parts that should stay mine with me.
- The draft run. Nod reads the next topic row from the sheet, writes the draft in my tone, generates the image, and creates the draft node on the site through the JSON:API. Then it verifies the draft exists and reports back with a link. It never publishes. Publishing stays mine, always.
- The Sunday ideas session. Nod reads what I published and drafted that week, proposes ten new rows, and waits for me to approve the list before it touches the sheet.
- The checks I do by hand. Confirm the draft landed, confirm the image attached, confirm nothing went live that shouldn't have. Nod does those reads and shows me the receipts.
What stays outside Nod:
- The topic list is still the source of truth.
- The tone guide still says how I write.
- The Key Auth key lives in Nod's vault, not in a prompt. If the vault can't hold it the way I need, the JSON:API step stays manual and Nod does the drafting only.
The two problems to solve first
Nod stops when the laptop sleeps. So a daily run needs an always-on machine or the self-hosted server option, which runs one person per install. A hosted version is on n8n's roadmap, but there's no date.
And it's a research preview from an experimental team, so I expect rough edges and frequent updates. That's fine for a trial. It's not fine for something I stop watching.
Final thoughts
The first run is one topic, one draft, with me watching every step. I'll measure the obvious things: did the draft land, did the image attach, did it stay a draft, and what did it cost. If that run's clean, the pipeline moves over for real. If it's not, I'll say so here.
Read next: Five AI tools I'm actually using every day