Since November 10, 2026, there has been an open alternative to Claude Cowork: Nod, an AI agent from n8n. Three things set Nod apart from ChatGPT Dots, Grok Bot and Claude Cowork: Nod is open source under Apache 2.0. You keep full control, because Nod runs on your own computer or server, files, memory and saved logins stay there, and the check that decides on approvals runs in your own Nod server. And you can use any model, according to n8n from more than 30 providers, through your own server or locally. With Dots, Grok Bot and Cowork, the vendor decides on the model and the platform. The downside of Nod: there is no flat-rate subscription. For most models you pay per use through the API; according to the docs you can sign in with a ChatGPT Plus or Pro subscription, while Claude only works with an API key. All vendor details as of November 15, 2026.
Claude Cowork works on Anthropic's servers and is included in every paid Claude plan, from Pro at $20 a month. ChatGPT Dots is OpenAI's always-on agent with its own computer in the cloud, included in Pro from $100 a month, but according to OpenAI not in the European Economic Area on that plan. Grok Bot is xAI's agent on a cloud computer in the US, included in The X Subscription from Premium at $30 a month. Claude in Chrome and Claude Code bring Claude onto your own computer, into the browser and the terminal. At OpenAI, the ChatGPT agent page now leads to ChatGPT Work, which runs locally in the desktop app or in OpenAI's cloud. And there is your own AI server with Claude Code and Slack, the way I built it for my AI employees. Which way fits depends on where your data may live, which model you want, which subscription you already have and whether the agent should keep working at night.
I work with fifteen AI employees, built my own Linux server for five of them, and used Cowork myself in the past. I installed the Nod server from source on my Mac on November 15, 2026, version 0.7.6, and set Nod against that setup here. Everything else about Nod comes from n8n's repository and docs, and everything about Anthropic, OpenAI and xAI from their help pages, docs and pricing pages, retrieved on November 15, 2026. Everything about AI agents is collected on my AI agents page.
Open source, full control, any model
Three points separate Nod from ChatGPT Dots, Grok Bot and Claude Cowork, and they weigh more than any single feature:
- Open source: Nod's code is public on GitHub under the Apache 2.0 license, and the software is free. You can read what it does and run it yourself.
- Full control: Nod runs on your computer or your server. Files, memory and saved logins stay there, and the check that decides on approvals runs in your own Nod server. The model provider still receives what each task needs, and app connections such as Gmail or Slack go through n8n and Composio by default. According to OpenAI, Dots works on its own computer in the cloud; according to xAI, Grok Bot runs only on Cursor's cloud computers in the US; Cowork runs on Anthropic's servers.
- Any model: According to n8n, more than 30 providers work, plus your own server with an OpenAI-compatible interface and local models through Ollama, LM Studio, vLLM or llama.cpp. According to xAI, Grok Bot has no model choice at all; Dots works with OpenAI models, Cowork with Claude.
The downside: Nod has no flat-rate subscription. For most models you pay per use through the API, directly to the model provider. The exceptions are a ChatGPT Plus or Pro subscription, which according to the docs you sign in with directly, and a local model on your own hardware. Claude only works with an API key, not with your Claude subscription. Nod only shows costs per task and a monthly limit for API models. Dots, Grok Bot and Cowork, by contrast, come in subscriptions with a fixed monthly price and included usage.
Five ways in one table
The "For whom" row is my assessment. The other rows come from the vendors' pages, and the last column from my own setup as I describe it in Claude Code on a Server: tmux, Remote Control.
Nod, ChatGPT Dots and Grok Bot
All three are agents with a computer of their own that keep working between your messages. The difference is who owns the computer and the model:
Nod: an open alternative
Nod gives the agent a computer of its own: a locked-down Linux workspace for commands and files and its own Chromium browser for websites, both as containers on the machine that runs Nod. With them, Nod fills in forms, signs in with saved logins and downloads what you need. Its README calls Nod an "open source personal agent with its own computer"; it is licensed under Apache 2.0 and free.
What stands out is where the rules live. A check in the Nod server decides which app actions, browser steps and sign-ins may run. It sees your words and the proposed action, but never the pages or emails Nod read. The server enters passwords, codes and cards into the website itself; the model never sees them. n8n also says plainly: the check is a model's judgement, and Nod is a research preview.
Three limits belong in any comparison: schedules only run while the computer running Nod is on. App connections go through a service hosted by n8n and through Composio by default. And usage data goes to n8n by default, which you can turn off in Settings → Privacy & data. How Nod works in detail is in Nod by n8n: What the AI Agent Can Do.
Claude in Nod: API key only
If you are looking for a Claude Cowork alternative, you probably already have a Claude subscription. That is exactly what you cannot use in Nod. The Nod docs say: "Anthropic's authentication conditions also require an API connection for Claude in Nod." Claude only gets into Nod through an Anthropic API key, and you pay per use on top of any subscription you have.
OpenAI is different: you sign in to Nod directly with a ChatGPT Plus or Pro subscription. The trade-off is transparency. With an API model, Nod shows under Usage what each task cost and stops at your monthly limit. With the ChatGPT subscription it only shows plan usage, without costs and without a limit.
For your model choice, n8n's tests are worth a look: in 240 test decisions of the approval check, according to n8n, GPT-6.1 Sol and Claude Opus 5.5 got every one right, while Claude Haiku 5.5 let 8 unsafe actions through.
Claude Cowork: tasks on Anthropic's servers
According to Anthropic, Cowork uses the same agentic architecture as Claude Code, just without a terminal. You describe an outcome, Claude plans, works in an isolated environment on Anthropic's servers and delivers documents, spreadsheets with working formulas or presentations. Since October 6, 2026, new tasks on Pro, Max and Ultra always run in the cloud; there is no longer a setting there to run them only on your own computer. For local folders, the browser and computer use, the desktop app has to be open. Scheduled tasks run in the cloud, with no device online.
For approvals, the new unified experience offers two levels: "Auto" is the default on Pro, Max, Ultra and Team; Claude keeps working and automated safety checks run before each action. "Manual" is the default on Enterprise and asks before actions. Claude always asks before permanently deleting files.
Cowork costs nothing extra but is only part of paid plans: Pro at $20 a month ($17 billed annually), Max at $100, Ultra at $200, Team from $20 per seat billed annually. According to Anthropic, multi-step tasks use more of your usage than a quick question. What else Cowork can do is in Claude Cowork: What It Does, Who It's For.
Claude in Chrome and Claude Code: Claude on your own computer
Claude in Chrome is Anthropic's extension for Google Chrome. Claude reads pages, clicks, types and fills in forms, in your own browser. It is available on every paid plan, but only in Google Chrome on the desktop, not in other Chromium browsers and not on mobile. The side panel runs as a Cowork session on Max, Ultra and Team, on Pro as the rollout reaches you, and starts in "Automatically approve" mode. More in Claude in Chrome: How to Use the Extension.
Claude Code is Anthropic's agent for the terminal, IDE, desktop app and web, included in every paid plan and sharing the same usage as the rest of the subscription. Its permission modes range from "Manual", which asks before most changes, through "acceptEdits" and "auto", where a second model reviews actions instead of you, to "bypassPermissions", which Anthropic reserves for isolated containers and VMs. For schedules, Anthropic names three ways: in the cloud without your computer but without local files, in the desktop app with your computer on, and with /loop inside an open session. When Claude Code fits better than Cowork is in Claude Code or Cowork: What For?.
ChatGPT Work: where ChatGPT agent leads
If you open OpenAI's ChatGPT agent page, chatgpt.com/features/agent, you are redirected to ChatGPT Work. OpenAI describes Work as a way to hand off a task with a clear outcome, such as a brief, a deck or a recurring update. Work uses files, plugins and approved tools, and in the desktop app also local files, apps and the browser. You follow progress, answer questions and approve important actions.
Tasks run locally in the desktop app or in the cloud. If your computer is off, an eligible task can continue in a cloud container, which then cannot reach your files. According to OpenAI, scheduled tasks that need local files require the computer to be on and the app running. Computer Use, meaning operating desktop apps, is available "in supported regions" on macOS and Windows according to OpenAI; ChatGPT asks before it uses an app.
According to OpenAI, Work is included in every ChatGPT plan: Free ($0) and Go ($8 a month) with GPT-6 Luna in the desktop app, subject to rollout; Plus at $20; Pro at $100, $200 or $500; Business at $20 per user billed annually from two users. Work and Codex share usage. I compared Work and Cowork directly in ChatGPT Work vs Claude Cowork: Delegating, and the pricing in detail in Codex Pricing and Usage Limits 2026.
ChatGPT Dots: OpenAI's always-on agent
According to OpenAI, Dots is an "always-on agent": you give it an ongoing responsibility, such as keeping a proposal current or organizing an event, and it keeps working between conversations, reports results and asks when a decision needs you. Your dot lives in the cloud with its own computer and browser, so it keeps working when your computer is off. If you want, you connect your own computer; it then has to be online with the ChatGPT app open. You reach it in ChatGPT on mobile, desktop and in the browser, in Slack, in Microsoft Teams and by call. On the model side, OpenAI describes a family of its own frontier models, now including GPT-6.1 Terra; your dot can use a different one for each part of the work.
Before an action affects your accounts or shares information, an automatic review decides whether the dot proceeds, asks you or hands the step over to you. You can add your own rules. You enter passwords into a private form that, according to OpenAI, sends them straight to the cloud browser without the model seeing them.
According to OpenAI, dots are included in Pro ($100, $200 or $500 a month), in Business Premium in supported regions and in eligible Enterprise, Edu and Healthcare workspaces. On Pro, dots are only available outside the European Economic Area, the UK and Switzerland. On Pro and Business Premium, conversations with your dot don't count toward your ChatGPT usage limits; tasks it hands to Work or Codex do. How Dots differs from Work is in ChatGPT Work vs Claude Cowork: Delegating, and Anthropic's counterpart in Claude Duet: What Anthropic's Agent Does.
Grok Bot: xAI's agent on a cloud computer
According to xAI, Grok Bot is a proactive AI teammate with access to a real computer in the cloud. Your Bots share a persistent cloud computer with a browser, filesystem and terminal, pick up where they left off and, if you want, repeat a task as a routine on a schedule. Several Bots can work together and hand work to each other. You reach them in the desktop app for macOS, Windows and Linux, in the iOS and Android app, on X and as a Team Bot in Slack.
Two points from xAI's own docs decide the comparison with Nod. First, the model: there is no choice. Grok Bot picks a model from SpaceXSI or a third-party provider for each task and does not show which one. Second, the location: Grok Bot runs only on cloud computers operated by Cursor, and they are in the US. According to xAI, running it inside your own network or on your own hardware is not supported today.
Approvals run through "Auto Review", a separate review model that assesses risky actions before they run, and through your own rules such as "Ask first" before any external email. According to xAI, the model never sees saved passwords. Grok Bot is included in The X Subscription from Premium at $30 a month ($25 billed annually, plus sales tax where required); Super costs $100, Ultra $200; Plus does not include it. Usage draws from the same weekly pool as Grok and Cursor. All plans and API prices are in Grok Pricing: Plans, Grok Bot and API, the comparison with Claude in Grok vs Claude 2026: Compared for Work, and what it means for customer data in Is Grok GDPR Compliant? A Business Checklist.
My own AI server with Claude Code and Slack
Your own AI server is not a product but a setup. In October 2026 I moved five of my AI employees onto a small Linux server in the Hetzner Cloud: content, office, marketing, SEO and website. Each has its own folder with a personnel file, rules and skills, and its own Claude Code session. They are steered through Slack: one bot and one channel per employee. Each bot only accepts jobs and approvals from my own Slack account, and tool approvals work by emoji reaction. Recurring jobs are schedules on the server with claude -p, and a Git sync saves the state every 15 minutes.
Next to Nod, three differences stand out. First, the model: my setup uses a Claude subscription through Claude Code, while Nod would require an API key for Claude. Second, the browser: a Linux server has no computer use and no Claude in Chrome, so when I built it, I left employees who click through web interfaces on the Mac. Nod brings its own browser in a container, on a server too. Third, approvals: in my setup they are rules in files and technical blocks in the settings, while in Nod the server checks every action beyond reading and drafting against your request. According to the docs, one Nod server belongs to one person; my setup keeps several roles separated on one server.
The server (Hetzner CAX21) costs 10.49 euros a month before VAT (as of October 2026), plus the Claude subscription. The real price is the setup: I built the server, credentials, Slack bots and schedules myself. The details are in Claude Code on a Server: tmux, Remote Control, remote control from the phone in Claude Code Remote Control from Your Phone, and who does what in My AI Workforce: Fifteen Employees, One Boss. How Nod and this setup compare on billing and cost, with a worked example, is in Nod vs Claude Code: AI Agents Compared.
Which way for whom
My assessment, as a rule of thumb:
- You have a Claude subscription and want to hand off office work: Cowork is already in your plan. Nod would need an extra API key for Claude.
- You want to choose your model or run it locally: of all the ways in this comparison, only Nod leaves the provider and model up to you, according to n8n more than 30 providers plus local models through Ollama or LM Studio.
- Your data should stay on your own hardware: Nod keeps files, memory and logins on your computer or server. The model provider still receives what each task needs, and app connections go through n8n and Composio by default.
- You already work in ChatGPT: ChatGPT Work is in the same account, and you can also sign in to Nod with your Plus or Pro subscription.
- You want an agent that reaches out on its own, without running anything yourself: ChatGPT Dots or Grok Bot, if the vendor's model and platform are fine with you. Dots is not available on Pro in the EEA, and Grok Bot runs on computers in the US.
- The agent should work at night: Cowork and Claude Code can schedule in the cloud without your device, and Dots and Grok Bot work in the cloud anyway. Nod needs your own server or a Mac mini for that.
- You want several roles running permanently by fixed rules: then your own server with Claude Code is worth the setup, even though it takes more work at the start.
Another open option that works through messengers is described with its risks in OpenClaw Alternative for Businesses. The direct comparison of Nod with OpenClaw and Hermes Agent is in Nod vs OpenClaw vs Hermes Agent Compared. Which agents run entirely on local models is in Running AI Agents Locally: What Works.
Frequently asked questions
What is the best Claude Cowork alternative?
There is no single best one; it depends on your data and your subscription. Nod is an open alternative with its own computer and a free choice of model, ChatGPT Work the alternative for anyone already working in ChatGPT, and your own server with Claude Code the alternative for always-on work by fixed rules.
Can I use my Claude subscription in Nod?
No. According to the Nod docs, Anthropic's conditions require an API connection for Claude in Nod. You need an API key and pay per use. A ChatGPT Plus or Pro subscription, by contrast, you can sign in with directly.
Is there a Nod subscription, or do I pay through the API?
There is no Nod subscription; the software itself is free. You pay the model provider, usually per use through an API key. According to the docs you can sign in with a ChatGPT Plus or Pro subscription; Nod then only shows plan usage, without costs and without a monthly limit. Claude only works with an API key. A local model has no cost per request. ChatGPT Dots, Grok Bot and Claude Cowork, by contrast, come in subscriptions with a fixed monthly price.
Where is ChatGPT agent today?
The page chatgpt.com/features/agent now redirects to ChatGPT Work. Work completes tasks with a clear outcome, locally in the desktop app or in the cloud, and asks before important actions. According to OpenAI it is included in every ChatGPT plan, from Free with GPT-6 Luna in the desktop app.
Does Nod keep running like Cowork when my laptop is closed?
No. Cowork keeps working on Anthropic's servers, Nod only while the computer running it is on. For always-on use, Nod runs on your own server or a Mac mini; a hosted version is on n8n's list for later.
Is Nod safer than Claude Cowork?
Both check actions automatically before they run, and both vendors say this check is not perfect. Nod enforces the check in its own server and keeps passwords and keys away from the model, but it is a research preview. How agents are protected against injected instructions is covered in Prompt Injection: Protecting AI Agents.
What to do next
Take the one task you would most like to hand off and write it down with a clear end and a clear limit: what should be done and what the agent may not do without you. Then check two things: which subscription you already have and where the data for this task is allowed to live. That usually tells you which way fits. How a task description like that turns into an AI employee with a personnel file is something I show on my hire AI employees page and in my community Claude Practitioners.
