Image: GitHub
Why it mattersA team wants one agent to work across websites, mail and calendar. Nod is an open option with 606 stars behind it.
Most AI agents that do real work for a person live in somebody else's cloud. The user's keys and the user's browser sit in a data centre. n8n released Nod on GitHub this week. It is an open-source personal agent that runs on the user's own Mac or Windows machine. The repository had 606 stars at the time of this item. It is under Apache 2.0. The company released version 0.6.0 on 9 November, 0.6.1 on 10 November, and 0.6.2 on 11 November.
Nod is a desktop app. It runs an AI agent on its own small computer on the user's machine. The user picks the model. OpenAI, Anthropic, Gemini, OpenRouter, Bedrock, Vertex AI and Azure are listed. The README says there are 20 more providers plus self-hosted options. A user with a ChatGPT Plus or Pro account can sign in with that. The app has a Mac build for Apple Silicon and Intel, and a Windows x64 build. Both are signed by n8n GmbH.
What tasks the agent does
The README names concrete jobs. Chase an unpaid invoice. Build a one-page brief for a sales call. File a subscription refund. Draft a Slack reply and wait for approval before sending. Watch a web page for a price change on a schedule. Turn a client request into a Linear task. Report weekly numbers. Book a team offsite.
Every one of these is a task on a real website or in a real app. The agent has its own browser and its own files. It asks the user before an action it has not seen before. It keeps a record of what it did and when.
The limits n8n names itself
n8n writes two limits into the README. The agent cannot run while the laptop is closed. A scheduled task works only when the machine is on. A user who wants the agent working overnight has to leave the computer powered up. A Mac mini or a self-hosted box also works. The second limit: the agent cannot make phone calls yet. The company lists that as a future feature.
The project is a research preview. n8n says in the README that the agent "can make mistakes" because the safety check is the model's own judgement. Approval before an unexpected action is the main line of defence. A user with no approval set up is trusting the model to call for help when it is unsure.
Why this matters for a team
The pattern n8n is testing is different from the one most agents use today. A hosted personal agent sits between the user and the websites it works on. The keys live in a vendor cloud. A local agent like Nod does the same job on the user's own machine. The keys stay there. So does the record.
The cost is that every machine needs the app, the models and the credentials set up. The gain is that nothing leaves the laptop unless the agent is told to send it.
A team looking at a personal agent for work has had few open options to compare against the hosted products. Nod is the first one built by a company with a mature automation business behind it. The code is on GitHub. The agent keeps a record on the machine. A team can read what it asked, what it did, and what it changed.
Source
n8n-io/nod on GitHub, with product details at nod.n8n.io.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
