Nod vs Claude Code: the biggest difference between Nod and your own AI server with Claude Code is how you pay. Nod, the open-source AI agent from n8n, usually pays for its model per use through an API key. Claude only works in Nod that way, while according to the docs you can sign in directly with a ChatGPT Plus or Pro subscription. Your own server with Claude Code instead uses a Claude subscription at a fixed monthly price, as long as its usage limits hold. In the cost example below, an agent with GPT-6.1 Sol or Claude Sonnet 5.5 costs about $14 a month for 20 tasks and about $210 for 300 tasks. In return, Nod gives you a free choice of model, its own browser and approvals checked by your own Nod server. The server with Claude Code carries several roles with fixed rules, but takes far more work to build. As of November 15, 2026.
I work with fifteen AI employees and built my own Linux server with Claude Code and Slack for five of them, on a Claude subscription rather than an API key. I installed the Nod server from source on my Mac on November 15, 2026, version 0.7.6. This article puts both ways side by side: my setup, as I described it in Claude Code on a Server: tmux, Remote Control, and Nod according to its repository and docs. Prices and terms come from the pricing and help pages of Anthropic, OpenAI and n8n, retrieved on November 15, 2026. Everything about AI agents is collected on my AI agents page.
The two ways
Nod: an agent with its own computer and the model you choose
According to n8n, Nod is an “open source personal agent with its own computer”: 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. The software is licensed under Apache 2.0 and is free. You start it as a desktop app, from source or on your own server, and talk to Nod in the app, in the browser, in Telegram or in WhatsApp.
You choose the model yourself. According to n8n, more than 30 providers work, plus your own server with an OpenAI-compatible interface and local models. When it comes to paying, there are three cases:
- API key, for example from OpenAI, Anthropic or OpenRouter: you pay per token directly to the provider. Nod shows under Usage what each task cost, sets a $200 monthly limit when you finish setup and checks in with you once a task has spent $5 since your last message.
- ChatGPT Plus or Pro: you sign in with your ChatGPT account, at no extra cost according to Nod. Nod then hides costs and the monthly limit. According to OpenAI, usage in other apps counts toward your plan's limits; you don't get any extra allowance from it.
- Local model through Ollama, LM Studio, vLLM or llama.cpp: according to the docs there is no cost per request, but you need suitable hardware.
How Nod works in detail is covered in Nod by n8n: What the AI Agent Can Do. All model options with prices and worked examples per option are in Nod Pricing: What the AI Agent by n8n Costs.
My setup: my own AI server with Claude Code on a subscription
I built my own AI server in October 2026 for five of my AI employees: content, office, marketing, SEO and website. As the machine I chose a small Linux server in the Hetzner Cloud, a CAX21 with arm64, 4 vCPUs and 8 GB of memory. This is how it is built:
- One folder and one session per employee. Each employee has its own folder with a personnel file, rules and skills, and its own Claude Code session in tmux, with Remote Control for the Claude app.
- Slack as the channel. One Slack bot and one channel per employee, connected through the open-source project claude-threads. Each bot is set up to accept tasks and approvals only from my own Slack account; tool approvals work by emoji reaction.
- Schedules without a screen. I set up recurring tasks as systemd timers that start Claude Code with
claude -p and a fixed permission mode.
- State in Git. For the working state I set up a Git sync every 15 minutes; credentials are stored encrypted in the repository.
For the model, I signed the server in with my Claude Team account, not with an API key. According to Anthropic, Claude Code is included in all paid Claude plans and shares its usage limits with the rest of the plan. According to Anthropic, Remote Control only works with a subscription; API keys are not supported there. Anything beyond the limits runs on additional usage credits, which you can turn on in your plan. According to Anthropic, the advertised limits for Pro, Max and Ultra assume ordinary, individual usage. For round-the-clock schedules in a company, a Team or Enterprise account or an API key is the better fit. All the details are in Claude Code on a Server: tmux, Remote Control, and steering it from your phone in Claude Code Remote Control from Your Phone.
Why Nod can't use your Claude subscription
The Nod docs say: “Anthropic's authentication conditions also require an API connection for Claude in Nod.” The reason is on Anthropic's side: signing in with a Claude account is intended exclusively for subscribers and for ordinary use of Claude Code and Anthropic's own applications. Anthropic allows third parties neither to build a Claude sign-in into their own apps nor to route requests through Free, Pro, Max or Ultra credentials. If you want Claude in a third-party tool like Nod, you need an API key and pay per use, on top of any subscription you already have.
OpenAI takes the opposite route. With “Sign in with ChatGPT”, eligible Go, Plus, Pro and Business subscribers can use their plan for AI requests in supported apps; according to the docs, Nod accepts Plus and Pro. Usage then draws from the same allowance you have in ChatGPT for Codex and Work.
Nod and your own AI server compared
The “For whom” row is my assessment. According to the docs, a Nod server belongs to one person; my setup separates several roles on one server. How to set up Nod on your own server or a Mac mini is covered in Nod Self-Hosted: AI Agent by n8n on Mac mini.
Cost example: API costs versus a subscription
What an agent costs via API depends on three things: how many tasks it does, how many model calls a task needs and how many tokens each call sends and receives. The formula:
Cost per month = tasks × calls per task × (input tokens × input price + output tokens × output price) ÷ 1,000,000
For the cost example I assume 10 model calls per task, each with 30,000 input tokens and 1,000 output tokens. The input is that large because, according to the Nod docs, Nod's instructions and tools alone take about 20,000 tokens before your first message. Per task that is 300,000 input tokens and 10,000 output tokens. The prices are the official list prices per million tokens from the OpenAI and Anthropic pricing pages:
Example for GPT-6.1 Sol: 300,000 × 2 ÷ 1,000,000 = $0.60 input plus 10,000 × 10 ÷ 1,000,000 = $0.10 output, $0.70 per task in total.
On the other side are the subscriptions: ChatGPT Plus costs $20 a month and can be signed in to Nod. Claude Pro costs $20 a month ($17 billed annually), Max $100, Ultra $200, and a standard seat on the Team plan $20 a month billed annually or $25 billed monthly; according to Anthropic, the Team plan is for 2 to 150 people. You use these Claude plans with Claude Code, not with Nod. With the numbers above, the threshold for GPT-6.1 Sol looks like this: up to just under 29 tasks a month the API is cheaper than a $20 plan, and from about 143 tasks it costs more than a $100 plan. Nod's default monthly limit of $200 would be reached after about 285 tasks with GPT-6.1 Sol and after just over 140 with Claude Opus 5.5.
What the example leaves out:
- Caching. The numbers assume no prompt caching. Repeated input from the cache costs only a fraction of the normal input price at both providers; the docs don't say whether or how much Nod uses it.
- Checks and background work. According to the docs, the approval check, check-in questions and background work such as memory upkeep call models on top and count toward the monthly limit.
- Subscription limits. Anthropic doesn't describe them in tasks but in relative terms, such as “5x Pro usage” on Max. According to Anthropic, how fast you reach a limit depends on the model, the length of your chats and the size of your files. Whether a plan carries 300 tasks a month only shows in your own use.
- Cheap isn't safe. 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. So the cheapest model is not automatically the best choice in Nod.
The assumptions are mine, not a measurement. Nod shows the real value per task under Usage. For comparison with Claude Code: for Claude Code via the API, Anthropic gives an enterprise average of about $13 per developer per active day and $150 to $250 a month. That's why a subscription is usually the more predictable route for daily work with Claude Code; what Pro, Max and Ultra include is covered in Claude Code Pricing: Plan or API?, and the API prices in detail in Claude API Key: Setup, Pricing, First Call.
When to use which
My assessment, as a heuristic:
- You want to try it out first and have ChatGPT Plus or Pro: Nod with your ChatGPT subscription. According to Nod it costs nothing extra, but it draws from the allowance you also use for Codex and Work.
- You have a few changing errands across websites with logins: Nod via API with a strong model and a low monthly limit. With few tasks a month, the bill in the cost example stays below a subscription, and Nod brings its own browser.
- You want to choose your model freely or run it locally: Nod. A server with Claude Code stays with Claude.
- You work a lot with Claude, every day, on files and code: Claude Code on a subscription. Nod would have to bill every request through the API.
- Several roles should work permanently by fixed rules: your own server with Claude Code, even though building it takes far more time. According to the docs, Nod is designed as one server per person.
- A team should share one agent: according to the docs, Nod isn't meant for that. For Claude Code in a team you take a Team or Enterprise plan; what that costs is covered in Claude Team: Cost and When It Pays Off.
The two don't rule each other out: Claude Code on a subscription for work on files and code, Nod for errands in the browser with its own monthly limit. How Nod compares with Claude Cowork, ChatGPT Dots and Grok Bot is covered in Claude Cowork Alternative: Nod Compared. What Nod sends to n8n and other services, and what you can switch off, is covered in Nod Privacy: What the AI Agent by n8n Shares.
Frequently asked questions
Why does Nod require an API key for Claude?
Because Anthropic limits signing in with a Claude account to Claude Code and Anthropic's own applications. According to Anthropic, third parties may not route requests through Free, Pro, Max or Ultra credentials. That's why Nod requires an API key for Claude, and you pay per use.
Is the API more expensive than a subscription?
It depends on volume. In the cost example, a task with GPT-6.1 Sol or Claude Sonnet 5.5 costs about $0.70. At 20 tasks a month that is $14, less than a $20 subscription; at 300 tasks it is $210. A subscription stays at its fixed price until its usage limits are reached.
Can I use Nod and Claude Code together?
Yes, side by side: Claude Code on a subscription for work on files and code, Nod for errands in the browser with its own monthly limit. As a model inside Nod, though, Claude Code gets you nothing. In Nod's list of model connections, Anthropic is listed with an API key only; there is no route through Claude Code or your Claude subscription.
What does your own AI server with Claude Code cost?
For my setup I chose a server at €10.49 net a month (Hetzner CAX21, as of October 2026), plus a data volume, backup storage and the Claude subscription. I signed the server in with a Team account; a standard seat there costs $20 a month billed annually, and according to Anthropic the Team plan starts at two people. The bigger item is the time to build it: you set up the server, credentials, Slack bots and schedules yourself.
Do I need my own server for Nod?
Not to get started. Nod runs as a desktop app on your computer, but schedules only run while the computer running Nod is on. For always-on operation, the docs name your own server or a Mac mini, with a starting size of two CPU cores, 8 GB of memory and 50 GB of disk.
What to do next
For one month, write down how often you would hand a task to an agent and which subscription you already have. If you're at a few errands a month and have ChatGPT Plus, the best move is to try Nod with that subscription right away; installation is covered in Install Nod: Set Up the AI Agent by n8n. If you already work with Claude every day, start with one AI employee in Claude Code on your own computer and only move to a server once folders, rules and approvals are in place. How I build my employees is something I show on my hire AI employees page and in my community Claude Practitioners.
