What an AI assistant is
ChatGPT, Claude and Gemini are chat assistants: you type a question in ordinary language and get an answer back, as if you’d asked a patient expert. Behind each one is a model, a program trained on a huge amount of text that predicts a good reply to what you write. These are the biggest and most capable models there are, and they run in the companies’ data centres, so all you need is a browser or a phone app.
For homelab work they’re the fastest teacher you can get. Paste an error message and ask what it means. Ask what each part of a command does before you run it. Ask for a settings file for an app, then ask it to explain every line. Things that once meant an evening of forum searching often take a few minutes.
Online or local: decide by what you’ll paste
Everything you type into an online assistant goes to the company’s servers and is stored there. That’s fine for an error message or a question about networking. It isn’t fine for passwords, private documents, other people’s details or anything you’d mind a stranger reading.
- Online assistants are cleverer, need no hardware and are free to start. Use them for learning, troubleshooting and anything that isn’t private.
- A local model, running on your own computer, keeps everything at home and works offline, but is less capable. See Local AI.
Most people use both, and choose by the question.
Why they’re useful, and why they’re wrong sometimes
An assistant doesn’t look facts up the way a search engine does. It writes the most likely-sounding answer from what it learned in training, which stops at a date, and it may search the web to fill gaps. That has two consequences for you.
First, it can be confidently wrong. It may invent a setting that doesn’t exist, mix up two versions of an app, or give a command for a different operating system. Second, it can’t see your machine. It only knows what you tell it, so the more exact your question, the better the answer.
| Give it this | Why it helps |
|---|---|
| The exact error message, copied rather than retyped | One word can change the answer |
| What you were doing, and what you expected | It needs the goal, not only the symptom |
| Your operating system and the app’s version | Instructions differ between versions |
| What you’ve already tried | So it doesn’t suggest the same thing again |
A screenshot works too. All three free plans can read pictures you upload.
What it needs
- An account. Each service asks you to sign up, with an email address or a Google, Apple or Microsoft login.
- The right age. Claude’s terms require you to be 18 or over. ChatGPT allows 13 to 17 year olds with a parent’s permission, and Gemini’s minimum age depends on your country and type of account.
- Nothing to install. A browser is enough, and each has free apps for phones.
- Patience with limits. Free plans cap how much you can use the best model, and the cap resets after a while. Paid plans raise the limits for a monthly fee.
How the free plans fit in
You don’t need to pay to get real use out of these. On the free plans of all three you can:
- Explain errors and logs. Paste the output and ask what went wrong and what to check first.
- Check a command before you run it. Ask it to explain each part, and what could go wrong.
- Write and check settings files, such as a Docker Compose file, then ask what each line does.
- Read documents for you. Upload a manual or a PDF and ask questions about it.
- Search the web for recent information and show where it came from.
- Teach you a topic step by step, such as how DNS works, and quiz you afterwards.
The options below cover the main assistants and the tools built for particular jobs, such as coding.
The easy way to start
- Open one in your browser, at chatgpt.com, claude.ai or gemini.google.com, or install its phone app.
- Sign up for a free account.
- Check the privacy settings first. In ChatGPT, open Settings, then Data controls, and turn off “Improve the model for everyone”. In Claude, open Settings, then Privacy, and turn off the switch for helping to improve Claude. In Gemini, turn off “Keep Activity” on the Gemini Apps Activity page. On free and personal plans, your chats can otherwise be used to train future models.
- Give each the same real problem from your own setup, such as an error you hit last week.
- Keep the one whose answers suit you, and use the others when it gets stuck.
ChatGPT’s temporary chats and Claude’s incognito chats aren’t used for training either way, which suits one-off questions.
If you’re comfortable in the terminal, coding agents such as Claude Code and OpenAI Codex, in the options below, can read and change files on your computer. They ask before acting, but give them a project you’ve backed up.
Questions people ask
Which one is best?
There’s no single answer, and they change every few months. All three handle homelab questions well. Try them on the same problem and keep the one whose explanations you understand most easily.
Is it safe to paste my config files?
Take out the secrets first: passwords, API keys, tokens, and anything that identifies your home, such as your public IP address or domain name. Replace them with placeholders like MY_PASSWORD. The rest of the file is usually fine to share.
Can I trust the commands it gives me?
Not blindly. Ask it to explain each part, compare with the app’s official documentation, and make one change at a time so you know which one broke something. Keep a backup of any file before you change it.
Should I pay for one?
Only when you keep hitting the free limits on something you do every week. One paid plan is enough for most people.
Can it set up my server for me?
Agents such as Claude Code can run commands on a computer, with your permission. For learning, it’s better to have the assistant explain each step while you do it yourself. Then you’ll understand your setup when something breaks.
