FixKit AI

Learn with Kit / Your AI toolkit

Your AI toolkit

Kit is just being Kit today. This is the topic Kit was born for.

One brain, every device — how to actually use your AIs

  1. Start with one assistant and one real chore

    Pick Grok Bot, Claude, ChatGPT, Grok or Gemini and give it a real job from your week — a letter, a bill to check, a plan for Saturday. Using it on real life is how it clicks.

  2. Your AI can follow you everywhere

    I talk to the same assistant on my computer, my phone and even in my truck. A question asked while driving becomes a note on my desk by the time I'm home.

  3. Bots can work as a team

    I run a small team of AI helpers — one for money, one for research, one for this website — with a manager bot that checks their work and only bothers me when a real decision is needed.

  4. Guardrails make AI safe to trust

    My voice 'decision AI' can only choose from a short list of safe actions. Approvals, read-only access and never sharing passwords are what make AI helpful instead of risky.

Your one-minute AI challenge

Open any AI assistant (Grok Bot, Claude, ChatGPT, Grok or Gemini) and paste this. Fill in the [brackets].

I'm new to AI. Ask me 5 quick questions about my week (chores, bills, family, work), then suggest the 3 easiest ways an AI assistant could save me time this week, with the exact words I'd type for each. Keep it simple and don't ask for any passwords or account numbers.

Fix-it-yourself guides for this

Chat assistants for everyday work

Which general AI assistant should I use day to day?

6 tools · 3-step recipe · copy-ready prompt

Build your own AI agent

I want an agent that uses tools and runs multi-step tasks.

6 tools

Run models on your own machine

I want AI that runs offline, free per-token, and private.

3 tools

Automate a workflow without code

When X happens in one app, do Y in another.

4 tools

The full guide

This is the home base for every other guide on FixKit AI. It shows you the main AI helpers (Grok Bot, Claude, ChatGPT, Grok, Gemini and Perplexity). It also covers free "open" AI you can run on your own computer, "agents" that can do chores for you, and how they all link together. I have used these tools every day since late 2022, and the biggest lesson is simple: you don't need to be technical. You need a few safe habits, a good question, and the courage to try.

👵 New to this? Start here

  1. Pick ONE assistant (Claude, ChatGPT, Grok or Gemini all have free versions) and install its app on your phone today.
  2. Ask it one real question from your life, like "Explain this letter to me in plain English." Then ask: "How sure are you? What should I double-check?"
  3. Make a rule and tape it by your computer: never paste passwords, full Social Security numbers, or bank or card numbers into any AI.
  4. Before you act on anything important (money, health, legal), check the answer on the official website the AI names, or ask a real professional.
  5. Once you feel comfortable, come back to this page and try one new idea a week: voice chat, a scheduled routine, or a second assistant to compare answers.

🎬 Real-life scenarios

Real situations — many from my own life, names removed — showing exactly what was asked and which AI did what.

The weekly report from a team of bots

👤 A dad of three who works in healthcare and runs a few side projects

The problem: He had several AI helpers doing different chores and was spending his evenings checking on each one.

  1. Used Claude to write a one-page "job description" for a chief-of-staff bot and for each worker bot (finance read-only, research, website check, social-post drafts).
  2. Set up the bots in Grok Bot with the rule: draft and report freely, but anything that sends, posts, buys or moves money waits for his yes.
  3. Asked the chief-of-staff bot: "Every Monday at 9 a.m., send me one short report: what each bot did, anything broken, and anything that needs my decision."
  4. Kept a shared notes file every bot reads first and updates last, so none of them repeat work.
  5. Spent the first two Mondays spot-checking each bot's claims before trusting the report.

What happened: His check-in time dropped from about an hour a night to about 15 minutes on Monday morning. The report caught two broken website links and one bot that had stopped running.

Grok BotClaudeShared notes

💡 One manager bot plus clear approval rules turns many helpers into one simple weekly report.

A question in the truck, the answer at the desk

👤 A busy parent who drives a Tesla Cybertruck

The problem: He had good ideas while driving but no safe way to write them down.

  1. While driving, pressed the voice button and asked Grok: "Give me three ideas for a simple weekend project with my kids, under $30."
  2. Listened to the short spoken answer, hands on the wheel, and said "remind me which one used cardboard."
  3. At home, opened the Grok app on his computer and asked again for the cardboard idea with a full supply list.
  4. Pasted the list into his family notes so his spouse could see it.

What happened: They built a cardboard marble run that Saturday for about $12 in tape and supplies. He never touched his phone while driving.

Grok (Tesla)Grok appNotes

💡 Use car voice for quick, simple questions; do the reading and deciding when you're parked.

Jev: a voice helper that can't do anything dangerous

👤 A dad of three who works from his home office

The problem: He wanted to open sites, switch tabs and change the volume by voice without worrying the AI would delete or send something.

  1. Wrote a plain-English list of safe actions (open apps and sites, switch tabs, scroll, volume) and a "never" list (delete, send, buy, move money).
  2. Asked Claude Code to build a small voice helper that may ONLY pick from the safe list and does nothing when unsure.
  3. Had Claude write a few hundred test phrases, including TV chatter and kids talking, and re-ran them after every change.
  4. Kept the helper limited to his own computer, opening sites in its own browser tab.

What happened: At first about 1 in 10 test phrases went to the wrong action; after several rounds of fixes, all of them passed. Mistakes are harmless: the worst case is that nothing happens.

Claude CodeMac speech recognitionJev

💡 A decision AI is safest when it can only choose from actions you approved.

Building a website with a team of AIs

👤 A specialty-pharmacy executive who is not a professional web developer

The problem: He wanted a free website that teaches people to fix their own problems with AI, but had little spare time to work on it.

  1. Used Claude Code as the "planner and builder": described each page in plain English and approved each change before it went live.
  2. Asked ChatGPT for homepage design ideas and picked the one he liked.
  3. Set up a Grok Bot helper as the "operator" to check every link each morning and report broken ones, without changing anything.
  4. Kept a status note that all three AIs read, so the planner knew what the operator found.

What happened: The site went from idea to live within a few days. The morning link check found a handful of broken links in the first week, all fixed the same day.

Claude CodeChatGPTGrok BotShared notes

💡 Give each AI one clear job (plan, design, operate) and one shared place to write down what happened.

One shared "brain" for five AI tools

👤 A dad of three who uses several AI assistants every day

The problem: Each AI forgot what the others had done, so he kept re-explaining the same projects.

  1. Made a notes folder with one "live status" page per project: goal, status, next steps, lessons.
  2. Told every assistant (Claude, Grok, and a free Hermes agent) to read the status page before starting and update it when done.
  3. Added a rule at the top: no passwords, account numbers or private health details in the notes.
  4. Once a week, asked Claude to tidy the notes and remove anything out of date.

What happened: He stopped re-explaining projects, and the AIs stopped redoing each other's work. A 10-minute weekly cleanup kept it accurate.

ClaudeGrokHermes AgentNotes app

💡 AI helpers work as a team when they share one set of notes.

Private help with a journal, on her own laptop

👤 A 72-year-old retired teacher with a 16 GB MacBook

The problem: She wanted help organizing years of personal journal notes but didn't want them sent to any company.

  1. Asked ChatGPT (with no private details): "Can a laptop with 16 GB of memory run a small AI model? Show the rough math."
  2. Installed LM Studio from its official website and downloaded a small 8B model (about 5 GB).
  3. Turned off her Wi-Fi and asked the local model: "Sort these notes by year and list the main topics."
  4. Checked a few summaries against the original pages.

What happened: It sorted about 200 notes in an afternoon. Answers were slower and simpler than ChatGPT, but nothing left her laptop and it cost $0.

ChatGPTLM StudioOpen-weight model

💡 For very private writing, a small model on your own computer is a real option.

Two assistants, one Medicare question

👤 A woman helping her 80-year-old father

The problem: She got two different answers from friends about whether a drug would be covered.

  1. Asked Claude: "In plain English, how do Medicare drug plans decide what's covered? Don't give advice."
  2. Asked Perplexity the same question and clicked its Medicare.gov source.
  3. Noticed the answers agreed on the basics but not on the details for his plan.
  4. Asked Claude to write 6 questions for his plan's customer service line, then called.

What happened: The 15-minute call answered all 6 questions. The drug was covered with a prior approval form his doctor sent that week.

ClaudePerplexityPhone

💡 Compare two AIs, check the official source, then let the real plan or professional give the final word.

Comparing AI prices before a hobby project

👤 A retired engineer building a news-summary helper

The problem: His agent program needed an AI key, and he didn't know which model would be affordable.

  1. Made one OpenRouter account and added $10 of credit with a $10 monthly limit.
  2. Opened the OpenRouter model list and picked one cheap model and one pricier one.
  3. Asked Claude to estimate the monthly cost of 100 short summaries with each, showing the math.
  4. Ran 5 test summaries with each model and compared them himself.
  5. Set the cheaper model as main and the other as a fallback.

What happened: The cheaper model was good enough and cost about 30 cents a month for his use; the fallback kicked in once during an outage.

OpenRouterClaudeHermes Agent

💡 Test a cheap model on your real task before paying for the most expensive one.

The voice-clone call

👤 A grandmother of four

The problem: She got a call that sounded exactly like her grandson saying he was in trouble and needed money fast.

  1. Remembered the family rule and asked for the secret code word; the caller dodged the question.
  2. Hung up and called her grandson on the number saved in her own phone.
  3. Later pasted a description of the call (no personal details) into ChatGPT and asked what scam signs it showed.
  4. Reported it at reportfraud.ftc.gov with her daughter's help.

What happened: Her grandson was safe at work. She lost $0, and the family now reminds each other of the code word every holiday.

ChatGPTPhoneFTC report site

💡 A family code word beats even a perfect fake voice.

Reading every page of a 12-page form

👤 A dad of three preparing papers for his tax preparer

The problem: A 12-page tax document arrived, and he was tempted to skim the first two pages.

  1. Covered account numbers, then uploaded all 12 pages to Claude.
  2. Asked: "Read every page. Tell me the page count first, then list anything on each page I might need to tell my preparer. Don't give tax advice."
  3. Checked that Claude's page count matched the paper.
  4. Turned the list into 5 questions for his preparer.

What happened: Page 9 had a detail he would have missed by skimming. His preparer confirmed it mattered and handled it.

ClaudePhone camera

💡 Ask the AI to confirm the page count, so you know it read the whole thing.

A weekly brief from favorite YouTube creators

👤 A busy parent who follows several investing and tech channels

The problem: He subscribed to dozens of channels and couldn't keep up with hours of video.

  1. Listed 6 favorite channels in a note.
  2. In Grok, set a weekly Automation: "Every Saturday at 8 a.m., summarize the main points from these channels' new videos this week, with the channel, date and link for each point. Mark anything that is opinion. If you can't read a video, say so. Don't give investment advice."
  3. Clicked through to watch only the 2 or 3 videos that looked most useful.
  4. Kept the briefs in a notes folder to spot themes over time.

What happened: His weekly video time dropped from about 6 hours to about 1. Each point came with a link so he could check it. A few weeks the brief said it couldn't read a video, which he liked better than a guess.

Grok AutomationsClaudeNotes

💡 Make every AI summary point back to the exact source, so you can check it.

Never missing a school event

👤 A mom of three juggling school and sports emails

The problem: Dates for field trips, games and picture day were buried in long newsletters.

  1. Each Sunday, pasted the week's school newsletters (no student ID numbers) into ChatGPT.
  2. Asked: "List every date, time and what we need to bring, in a table sorted by date. If a date is unclear, say so."
  3. Checked each item against the original email and added them to the family calendar.
  4. Shared the calendar with her spouse so both phones got reminders.

What happened: The Sunday routine takes about 10 minutes. Over a semester, they stopped missing permission slips and "bring $5" days.

ChatGPTFamily calendarEmail

💡 AI is great at pulling dates out of long messages; you still double-check and click "add."

Testing whether a "winning strategy" was just luck

👤 A dad of three curious about trading research

The problem: An AI helped him find trading rules that looked amazing on past data, and he wondered if it was real.

  1. Asked Claude: "How do I check if these results could be pure luck? Explain a coin-flip test in plain English."
  2. Asked Claude Code to build the test: run thousands of random "coin-flip" strategies on the same data and show how often luck alone looked as good.
  3. Compared his best rule against the random ones, after including trading costs.
  4. Wrote the lesson in his shared notes so no AI would "rediscover" the same idea.

What happened: Most of his "edges" did no better than random coin flips once costs were included. He saved himself from risking real money on luck.

ClaudeClaude CodeShared notes

💡 AI can find patterns in anything, so always ask it to test whether the pattern is just luck.

From the mailbox to a to-do list, without retyping

👤 A dad of three who hates paper mail

The problem: Bills, renewal notices and school forms piled up on the counter, and due dates got missed.

  1. Set his desk scanner to save every scan into one "Inbox" folder in his cloud drive.
  2. Each evening, asked Claude (connected read-only to that folder): "For each new scan, tell me what it is, any due date, and what I need to do. Don't give financial or tax advice."
  3. Asked Claude to write the due dates as a short list, then pasted it into his "Renewals and deadlines" document himself (Claude only had read access).
  4. Checked the list on his phone the next morning and moved real deadlines into the family calendar himself.
  5. Covered or blurred account numbers on anything he later shared with his tax preparer.

What happened: About 15 scans a week now take him about 10 minutes to review. In the first two months he caught a car registration renewal and an insurance form two weeks early.

Desk scannerCloud driveClaudePhone calendar

💡 Scanner to folder to AI to calendar: each tool does one small job, and you approve the last step.

Private medical papers, read on her own computer

👤 A widow sorting her late husband's medical and insurance papers

The problem: She had a shoebox of hospital statements and wanted help making sense of them, but didn't want private health details sent to any company.

  1. Asked Claude a general question with no personal details: "What should I look for on a hospital statement and an insurance Explanation of Benefits?"
  2. Had her nephew install LM Studio on her 32 GB Mac mini and download a small open model, about 8B in size.
  3. Scanned the statements into a folder, turned off Wi-Fi, and asked the home model: "List each statement's date, provider, amount billed and amount I may owe, in a table. Don't give medical, legal or financial advice."
  4. Checked every row against the paper, because small models make more mistakes.
  5. Took the table and Claude's checklist to the hospital's billing office.

What happened: About 40 statements became one 2-page table in an afternoon. The billing office found one duplicate charge of about $300 and removed it.

ClaudeLM StudioOpen-weight modelScanner

💡 Ask the big online AI general questions; keep the private papers on a home model, and check its work.

A bakery's order-check bot, with a manager watching it

👤 A small-bakery owner who takes online orders

The problem: Custom cake orders came in through her website and email at all hours, and she sometimes missed one until the day before.

  1. Showed a Grok Bot helper once how she checks new orders, so it saved that as a routine.
  2. Told it: "Every 2 hours from 8 a.m. to 8 p.m., check for new orders and send me a 3-line summary on my phone. Never reply to customers, change prices or take payments."
  3. Set up a second "manager" bot to check at 10 a.m. each day that the order bot actually ran, and to tell her if it missed a check.
  4. Each Monday, compared the bot's summaries with her real order list for 10 minutes.

What happened: In the first month the bot flagged about 60 orders and she missed none. The manager bot caught one morning when the order bot didn't run, and she checked by hand that day.

Grok BotPhoneWebsite and email

💡 A routine bot does the checking, a manager bot checks the checker, and you keep every customer-facing action.

A safe voice helper for Grandpa's computer

👤 A grandfather with arthritis who finds the mouse painful, and his tech-savvy granddaughter

The problem: He wanted to video-call family, play his radio station and read the news without clicking, but the family worried a voice AI might do something unwanted.

  1. The granddaughter wrote a fridge-note list with him: only "call [family member] on video," "play the radio," "open the news," "volume up/down," and "stop."
  2. She asked Claude Code: "Build a tiny voice helper for this Mac that may ONLY do these 6 actions. If it is unsure, do nothing. Explain the plan first and wait for my OK."
  3. Asked Claude to write 30 test phrases, including TV noise and phrases it should ignore, and ran them together.
  4. Set the video-call action to show the call screen so Grandpa still presses "Call" himself.
  5. Checked in by phone after a week to see which phrases he actually used.

What happened: After one weekend of setup he used it about 10 times a day. Twice it didn't understand him, and both times it simply did nothing, which is exactly what they wanted.

Claude CodeClaudeMac speech recognitionPhone

💡 Copy the Jev idea: a short safe list, "unsure means do nothing," and a person still presses the final button.

Meet the assistants

An AI assistant is an app you talk to or type to, like texting a very well-read helper. These five are the big ones. Each has a free version and paid plans, and the plans change often, so always check the current plan page.

Claude (made by Anthropic)

Claude is an AI helper that is very good at reading long papers and explaining them simply.

📖 Example

Dorothy gets a 20-page letter from her insurance company. She covers her member ID with a sticky note, takes a photo of each page, and asks Claude what it means. Claude tells her in five sentences that her plan changes in January and lists the two dates she must remember. She double-checks both dates on the letter itself.

Claude is strong at careful reading, writing and long documents. You can hand it a 30-page letter or contract and ask for the key points in plain words. The free plan includes web search, memory across chats, and "connectors" (links to your other apps, like your calendar). Paid plans add more use, plus tools like Claude Code (for building software) and Claude in Chrome (it can work inside your web browser). I use Claude as my main "planner" for big projects.

🤖 How AI helps: Upload a long document and ask Claude to summarize it, list every deadline, and flag anything it is unsure about.

✅ Your first step: Open the Claude app or website, type "Explain what you can and can't do, in five bullet points," and read the answer.

Try this prompt
I'm new to AI. Read the document I attached and give me: (1) a 5-sentence plain-English summary, (2) every date or deadline, (3) anything that sounds like it needs a professional. Don't give legal, tax, medical or financial advice. Stop after those three lists and ask if I want more.

Claude plans (Anthropic) ↗ · Claude app (App Store) ↗

ChatGPT (made by OpenAI)

ChatGPT is the most famous AI helper; it can chat, write, make pictures and talk out loud with you.

📖 Example

Frank wants to plan his granddaughter's 10th birthday. He presses the voice button and talks for two minutes about the budget ($200), the guest list and the park. ChatGPT turns it into an 8-item checklist and a short party invitation he can text to parents.

ChatGPT is the best-known assistant and a great all-rounder: everyday questions, writing, pictures, voice chats and research. It has a free plan and several paid levels (OpenAI lists Go, Plus and Pro, plus business plans). Its "do the work for you" features have changed names more than once. Today OpenAI's help center calls the feature for longer, multi-step jobs ChatGPT Work, on paid plans. Check what your own plan shows.

🤖 How AI helps: Use voice mode to talk a problem through out loud, then ask ChatGPT to turn the talk into a short to-do list.

✅ Your first step: In the ChatGPT app, tap the voice button and ask, "What's one thing you could help me with this week?"

Try this prompt
Help me plan [a task, e.g. a family reunion for 25 people]. Ask me up to 5 questions first, one at a time. Then give me a simple checklist with no more than 10 items. Don't book, buy or contact anyone. Stop after the checklist.

ChatGPT app (App Store) ↗

Grok Bot and Grok (made by xAI)

Grok Bot is the one most people should try: a team of AI helpers that do chores for you while you're away. Grok is the chat app it lives in — on your phone, computer, the X app and some Tesla cars.

📖 Example

Ray drives a Tesla that has Grok. On the way to the store he asks out loud, "What's a quick dinner I can make with chicken, rice and frozen peas?" Grok answers by voice in about 10 seconds. At home he opens the Grok app on his phone to get the full recipe written out.

Grok is xAI's assistant. You can use it in its own app on your phone or computer, on the web at grok.com, inside the X (Twitter) app, and by voice in many newer Tesla cars. Grok can also run Automations: questions you schedule to run by themselves, like a morning news brief. Grok Bot is a newer beta product: a team of AI helpers that keep working while you're away and check in when they need your OK. (More on Grok Bot in the agents section below.)

🤖 How AI helps: Set up one Grok Automation, like "Every weekday at 7 a.m., give me 3 headlines about my town, with links," and see if it saves you time.

✅ Your first step: Install the Grok app (or go to grok.com) and ask it: "What are Automations and how would I set up a simple one?"

Try this prompt
Every weekday at [7:00 a.m.], give me the 3 most important news stories about [my city or a topic], one sentence each, with a link to the original source. If you can't find a reliable source for a story, leave it out. Stop after 3.

Grok ↗ · Grok app (App Store) ↗ · Grok Automations (scheduled prompts) ↗ · xAI: Introducing Grok Bot ↗ · Not a Tesla App (news): Grok rollout and requirements ↗

Gemini (made by Google)

Gemini is Google's AI helper, and it works best if you already use Gmail and other Google apps.

📖 Example

Linda has 40 emails about her church bake sale. On a paid Google plan, she asks Gemini in Gmail to summarize who is bringing what. It gives her a list of 12 names and items in seconds. She spots one mistake and fixes it before sending the list out.

Gemini is Google's assistant. It fits naturally if you already use Gmail, Google Docs, Google Photos or an Android phone. The free version handles chat, image making and research. Google's paid plans (Google AI Plus, Pro and Ultra) add more use, extra storage and Gemini inside apps like Gmail and Docs. Check the current plan page because features move between tiers.

🤖 How AI helps: If you live in Google apps, ask Gemini to help draft an email or summarize a long thread, then read it before you send anything.

✅ Your first step: Go to gemini.google.com, sign in with your Google account, and ask it to explain one email you didn't understand (cover any account numbers first).

Try this prompt
Here is an email I don't understand: [paste it, with account numbers and personal details removed]. Explain it in 3 plain-English sentences. Tell me if it asks me to do anything and by when. If it looks like a scam, say why. Don't give legal or financial advice. Stop there.

Google Gemini ↗ · Gemini plans (Google) ↗

Perplexity (the "answer engine")

Perplexity is like Google that writes you a short answer and shows exactly which websites it read.

📖 Example

Walt wonders whether his town's leaf pickup has changed. Perplexity gives a 3-sentence answer with two links, one to the town website. He clicks it and sees the new date, Nov. 4, posted by the town itself.

Perplexity works like a search engine that writes you an answer and shows the web sources it used, with numbered links. That makes it easy to check where a fact came from. It has a free version and a paid Pro plan; check the current plan page for what each includes. I use it when I want to see sources first and opinions second.

🤖 How AI helps: When an answer matters, ask Perplexity the same question and click at least two of its sources to see if they really say what it claims.

✅ Your first step: Ask Perplexity one question you already know the answer to, and click its sources to see how it works.

Try this prompt
What does [an official agency, e.g. Medicare] say about [my question]? Use only official or well-known sources and show each link. If sources disagree, tell me. Don't give personal advice. Stop after one short answer.

Perplexity ↗ · Perplexity Help Center ↗

Which one should I start with? (and how to ask well)

It doesn't matter much which AI you pick first; what matters is asking clearly and checking important answers.

📖 Example

Betty asks both ChatGPT and Claude, "Is it safe to take ibuprofen with my blood pressure pill?" One says "maybe not," the other says "ask your pharmacist." Because the answers don't match, she calls her pharmacist, who gives her a clear answer in two minutes.

Honestly, any of them. The free versions are good enough to learn on. What matters more is how you ask: say who you are, what you need, what to avoid, and when to stop. A trick I use every day: ask two different assistants the same important question. If they disagree, that's your signal to check the official source or ask a person.

🤖 How AI helps: Paste the starter prompt below into any assistant and fill in the brackets; it teaches the AI how you like to be helped.

✅ Your first step: Copy the starter prompt on this card into the AI you already have and try it on one small task.

Try this prompt
I'm a beginner and a little nervous about AI. Please talk to me in plain English, no jargon. I need help with [my task]. Ask me up to 3 questions first. Then give me no more than 5 steps. If you're not sure about something, say so and tell me where to check. Don't give legal, tax, medical or financial advice. Stop after the 5 steps.

Big-company AI vs. free "open" AI

An AI's "brain" is called a model. There are two kinds. Frontier models are the newest, most powerful ones; big companies run them for you online. Open-weight models are free to download, so anyone can run them on their own computer.

Frontier models: rented power

A frontier model is the biggest, smartest kind of AI, and it lives on a company's computers, not yours.

📖 Example

Hank asks ChatGPT to rewrite his resume. The words go over the internet to OpenAI's computers, which do the thinking and send back a new version in 15 seconds. It's fast and smart, but Hank decides not to include his home address and phone number until the final copy.

Claude, ChatGPT, Grok and Gemini run on frontier models: huge AI brains in company data centers. You get the most capable answers and the easiest apps, but your questions travel to the company's computers, and heavy use costs money. Each company has its own privacy settings; it's worth spending five minutes reading them.

🤖 How AI helps: Ask your assistant: "Where do my chats go, are they used to train you, and how do I turn that off?" Then confirm in the app's actual settings.

✅ Your first step: Open the settings in the AI app you use and find the page about data, privacy or "training." Just read it; don't change anything yet.

Try this prompt
In plain English, explain how [this AI app] handles my chats: are they saved, are they used to train the AI, and where in the settings can I change that? If you're not sure, say so and point me to the company's official help page. Stop after 5 bullet points.

Claude plans (Anthropic) ↗ · Gemini plans (Google) ↗

Open-weight models: AI you can download

An open-weight model is an AI brain a company lets you download for free and run on your own machine.

📖 Example

Grace downloads Google's Gemma, a smaller open model, onto her laptop using a free app. She asks it to tidy up her recipe notes while her Wi-Fi is off. Nothing leaves her laptop, and it costs her $0, though answers take a little longer than ChatGPT.

Open-weight means a company shares the finished AI brain as a file. Anyone can download it and run it. Well-known families include Meta's Llama, Mistral's open models, Alibaba's Qwen, DeepSeek, Google's Gemma, and Nous Research's Hermes models. Each comes with its own license (rules for use), so check it. They are usually a step behind the very top frontier models, but they can be free, private and work without the internet.

🤖 How AI helps: Ask an assistant to explain a model's license in plain English before you use it for anything beyond personal tinkering.

✅ Your first step: Visit huggingface.co/models and scroll for one minute to see how many free AI models exist. You don't need to download or sign up for anything.

Meta Llama ↗ · Mistral AI docs (open-weight models) ↗ · Qwen on Hugging Face ↗ · DeepSeek on Hugging Face ↗ · Google DeepMind: Gemma ↗ · Nous Research Hermes models on Hugging Face ↗

The tradeoffs in plain words

Big online AI is smarter and easier; home AI is more private and free but needs a good computer.

📖 Example

Sam is writing a family history from old letters that mention medical details. For those private letters he uses a free open model on his own computer. For general research about his town's history he uses Claude. His only bill is Claude's monthly plan; the home model costs nothing.

Cost: frontier apps are cheap or free to start, but heavy use can mean a monthly bill; open models are free but may need a stronger computer. Privacy: open models on your own computer keep everything at home; frontier apps send data to the company. Ability: frontier models are usually smarter and easier; open models need a bit of setup. Many people, including me, use both.

🤖 How AI helps: Describe your situation (budget, computer, privacy needs) and ask an assistant to lay out the pros and cons of each path without telling you which to choose.

✅ Your first step: Write down one task you'd only trust to AI if it stayed on your own computer, and one task where you don't mind.

Try this prompt
Here is my situation: [my computer, e.g. a 2021 MacBook with 16 GB of memory], [my budget], [how private my tasks are]. Lay out the pros and cons of using a big online AI vs. running a free open model at home, in a simple two-column table. Don't recommend a product. Stop after the table.

Nous Research's Hermes models

Hermes is a family of free AI brains from a research group called Nous Research.

📖 Example

Ken, a retired engineer, reads the Hermes 4 page on Hugging Face and sees a 14-billion-size version. He pastes the page into Claude and asks, "Could this run on my computer with 32 GB of memory?" Claude explains the size math and tells him to double-check with the app he plans to use.

Nous Research is an open AI research group whose stated mission is open access to intelligence. Its Hermes 4 family of open models comes in several sizes (from about 14 billion to 405 billion parameters) and is posted on Hugging Face. The same group makes Hermes Agent, a free assistant program covered in the agents section below.

🤖 How AI helps: Ask an assistant to explain what a model card on Hugging Face says (size, license, intended use) in plain English.

✅ Your first step: Open huggingface.co/NousResearch and read just the names of the models listed; notice the numbers like 14B and 70B (we explain them next).

Nous Research ↗ · Nous Research Hermes models on Hugging Face ↗

Running AI on your own computer

You can run an open model at home with a free app. The main question is memory: the AI brain has to fit in your computer's memory (RAM), or in the graphics card's memory (VRAM). Here's how to tell if yours is big enough.

What "7B" or "70B" means

The "B" number tells you how big the AI brain is; bigger is smarter but needs more computer memory.

📖 Example

Think of an 8B model as a helpful high-school student and a 70B model as a college professor. Rita's 8B model writes a nice thank-you note in 20 seconds. When she asks it a tricky tax question, it confidently makes up a rule, which is why she checks big questions with official sources.

The number with a B is how many billions of "parameters" (tiny dials the AI learned) the model has. Bigger usually means smarter but hungrier for memory. Common sizes are 7B or 8B (small), 14B (medium), 32B (large) and 70B (very large). Most people start with a 7B or 8B model.

🤖 How AI helps: Ask an assistant: "In plain English, what can an 8B model do well, and where will it struggle compared to ChatGPT?"

✅ Your first step: Find out how much memory your computer has: on a Mac, Apple menu > About This Mac; on Windows, Settings > System > About.

Hugging Face model library ↗

The memory rule of thumb

Compressed AI models need roughly half a gigabyte of memory for every billion "dials," plus a little extra.

📖 Example

Joe's laptop has 16 GB of memory. An 8B model at 4-bit needs about 8 × 0.55 = 4.4 GB plus a couple GB extra, so about 6–7 GB: it fits comfortably. A 70B model would need about 40+ GB, so it won't fit on his laptop.

Home AI apps usually use "compressed" models. Compressing (experts call it quantizing) shrinks each dial from 16 bits to about 4 bits. The model then fits in much less memory and loses only a little quality. Rough rule: a 4-bit model needs about 0.5 to 0.6 GB of memory for every billion dials, plus a few GB extra. So, roughly: 7B or 8B needs 5–6 GB, 14B needs 9–10 GB, 32B needs about 20 GB, and 70B needs 40–45 GB. These are estimates; your app shows the real file size.

🤖 How AI helps: Tell an assistant your computer's memory and ask which model sizes would likely fit at 4-bit, and to show its math.

✅ Your first step: Find your computer's memory and use this rough starting guide for 4-bit models: 8 GB → small 3B–4B models; 16 GB → 7B–8B (maybe 14B); 32 GB → 14B (maybe 32B); 64 GB or more → up to 70B. It's only a guess; the app will show the real file size.

Try this prompt
My computer has [16] GB of RAM and [no graphics card / a graphics card with 8 GB VRAM / an Apple M-series chip]. Using the rough rule that a 4-bit model needs about 0.5–0.6 GB per billion parameters plus a few GB of overhead, which sizes (8B, 14B, 32B, 70B) might fit? Show the math, say these are estimates, and don't recommend anything to buy. Stop after a short table.

Hugging Face: Quantization overview ↗ · Hugging Face: GGUF (compressed model files) ↗ · Hugging Face: Model memory estimator ↗ · LM Studio system requirements ↗

Macs, graphics cards, and plain computers

Newer Macs and PCs with a good graphics card run home AI well; older plain computers can run small ones slowly.

📖 Example

Maria has a Mac mini with 32 GB of memory. A 14B model answers her in a few seconds. Her brother's 8-year-old PC with no graphics card runs a small 3B model, but each answer takes about a minute, which is fine for him for writing letters.

Newer Macs (M1 chip or later) are good at home AI because the main chip and graphics share one big pool of memory, called unified memory. LM Studio recommends 16 GB or more, and it doesn't support older Intel Macs. On a Windows or Linux PC, a graphics card with its own memory (VRAM), such as one from NVIDIA, runs AI fastest. LM Studio recommends at least 4 GB of VRAM and 16 GB of RAM. A computer with no graphics card can still run small models, just slowly. Always leave some memory free for the computer itself.

🤖 How AI helps: Ask an assistant to explain your exact computer's specs to you in plain words, and what that means for running home AI.

✅ Your first step: Look up your computer's chip and memory (Mac: About This Mac; Windows: Settings > System > About) and write both down.

LM Studio system requirements ↗

Ollama and LM Studio: the easy apps

LM Studio and Ollama are free apps that let you download an AI brain and chat with it on your own computer.

📖 Example

Ed installs LM Studio from lmstudio.ai, searches for a small Llama or Gemma model, and clicks download (about 5 GB, 10 minutes on his Wi-Fi). Then he turns his Wi-Fi off and asks it to summarize his garden journal. It works with no internet at all.

Two free apps make home AI about as easy as installing any program. LM Studio has a friendly window where you search for a model, click download and start chatting. Ollama is a small app with a model library, and many other tools can plug into it. Both have versions for Mac, Windows and Linux (LM Studio needs a newer M-series Mac). Download them only from their official websites.

🤖 How AI helps: If you get stuck installing, copy the exact error message into an assistant and ask what it means in plain English.

✅ Your first step: Visit lmstudio.ai or ollama.com and read the home page; don't install yet, just see what it says your computer needs.

Try this prompt
I'm a beginner. I want to try [LM Studio / Ollama] on my [computer, e.g. 2022 MacBook Air with 16 GB memory]. Give me the steps to install it from the official website and download one small model, in no more than 8 numbered steps. Tell me which steps need the internet. Don't suggest any paid software. Stop after the steps.

LM Studio ↗ · Ollama ↗ · Ollama model library ↗ · Ollama docs ↗

OpenRouter: one key, many AI brains

Once you start using AI in other tools, you may want to try different models without signing up everywhere. OpenRouter is a service that gives you one account and one key for hundreds of models.

What OpenRouter does

OpenRouter is like one universal remote that works with hundreds of different AI brains.

📖 Example

Tom's home agent program asks for an AI key. Instead of signing up with five companies, he makes one OpenRouter account, adds $10 of credit, and pastes in one key. Now he can switch his agent between models with a dropdown.

OpenRouter gives you one place to reach many AI models from many companies, with one bill and one usage report. You get an "API key" (a secret password for software, not for people) that you paste into apps or agents that ask for one. OpenRouter says it charges the same model prices as the AI companies do. It adds a fee when you buy credits, so check its FAQ for the current fee.

🤖 How AI helps: Ask an assistant to explain what an API key is and how to store it safely before you create one.

✅ Your first step: Open openrouter.ai/models and sort by price just to see how much costs can differ between models.

OpenRouter ↗ · OpenRouter FAQ (pricing, fees, privacy) ↗ · OpenRouter quickstart ↗

Comparing prices

Different AI brains charge very different prices, and OpenRouter shows them side by side.

📖 Example

Ann wants to summarize 100 news articles a month. One model would cost about $2 for that and another about 20 cents. She tries the cheaper one on five articles first, sees the summaries are good enough, and keeps it.

AI services charge by "tokens" (small chunks of words, roughly three-quarters of a word each), usually priced per million tokens. OpenRouter's model list shows the price for each model side by side, and some models have free versions with daily limits. Prices change often, so check the live list instead of trusting any number you read online.

🤖 How AI helps: Paste two model prices into an assistant and ask it to estimate the cost of a typical task, showing the math.

✅ Your first step: On OpenRouter's model list, find one model marked free and one paid model and compare their listed prices.

Try this prompt
Model A costs [$X] per million input tokens and [$Y] per million output tokens. Model B costs [$X2] and [$Y2]. If I summarize [20] articles of about [1,000] words each and get back [200]-word summaries, estimate what each would cost. Show the math and say it's an estimate. Stop after the two totals.

OpenRouter model list and prices ↗

Fallbacks: a backup when one model is down

A fallback is a backup AI that steps in if your first choice isn't working.

📖 Example

Jim's morning-news agent uses one model. One day that company has an outage. Because he listed a second model as a backup, the news brief still arrives at 7 a.m., written by the backup model.

If one AI company's service has an outage, your tools can break. OpenRouter can automatically try another provider, and you can list backup models to use if the first choice fails. For a family or small-business setup, that means fewer "it stopped working" mornings.

🤖 How AI helps: Ask an assistant to explain the OpenRouter fallback page in plain English and how it would apply to your one tool.

✅ Your first step: Write down which AI tool you'd miss most if it stopped working for a day; that's the one that would need a backup.

OpenRouter: model fallbacks ↗

Guard your key like a credit card

An API key is like a credit card number for AI apps, so keep it secret and keep the balance small.

📖 Example

Pat accidentally pastes her key into a public help forum. Within an hour she notices $8 of charges she didn't make. She deletes that key, makes a new one, and sets a $10 monthly limit so a mistake can never cost more.

Anyone with your API key can spend your credit. Never paste it in a chat, email or public website. Set a spending limit, keep only a small balance, and delete (revoke) any key you think leaked. OpenRouter says it does not log your prompts by default, but read its privacy settings yourself.

🤖 How AI helps: Ask an assistant for a simple checklist for storing and limiting an API key; never paste the key itself into the chat.

✅ Your first step: If you've ever made an AI key, sign in to that service and open its usage or billing page. Look for charges you don't recognize. Don't copy the key anywhere.

OpenRouter FAQ (pricing, fees, privacy) ↗

Agents: AI that does chores for you

A chatbot answers questions. An agent can actually do things: click around websites, use apps, run routines on a schedule, and report back. That's powerful, so agents need rules: what they may do, and when they must ask you first.

What an agent is (and isn't)

An agent is an AI that doesn't just talk; it can use apps and websites to finish a job for you.

📖 Example

Carol asks an agent to find three plumbers near her with good reviews and put their numbers in a list. The agent searches, reads reviews and makes the list in 5 minutes. It does not call or book anyone, because she told it to only make the list.

An agent is an AI that is given tools (a web browser, email, files, apps) and a goal, and then takes steps on its own until the job is done or it needs your OK. Good agents ask before anything you can't undo, like sending, buying or deleting. They still make mistakes, so start them on low-stakes chores and check their work.

🤖 How AI helps: Before giving an agent a chore, ask it to describe its plan step by step and say which steps it will need your approval for.

✅ Your first step: Write down one boring computer chore you do every week, in one sentence (for example, "Check the library website for my holds"). That's a future job for an agent.

Try this prompt
Before you do anything, write out your plan to [task] as numbered steps. Mark every step that sends, buys, deletes, or shares something with [ASK FIRST]. Do not start until I reply "go." Never enter passwords or payment details. Stop after showing the plan.

Simon Willison: The lethal trifecta for AI agents ↗

Claude Code and Claude in Chrome

Claude Code builds and fixes software for you, and Claude in Chrome can click around websites for you while you watch.

📖 Example

Mike runs a small hobby website. He asks Claude Code to "find every broken link and show me a list before fixing anything." It finds 7 broken links in two minutes, shows him the list, and fixes them only after he types "yes."

Claude Code is Anthropic's helper for building and fixing software. It can read and change the files in a project, after you approve. Claude in Chrome is an add-on for the Chrome web browser, on paid Claude plans, that can click around websites for you. Anthropic says you approve each website one at a time and that Claude stops before purchases and money actions. It also warns that some websites hide instructions meant to trick AI, so watch what it does and keep it away from banking and health-record sites. I used Claude Code to build this very website.

🤖 How AI helps: Ask Claude Code to explain a project's files to you before you let it change anything.

✅ Your first step: Read Anthropic's Claude in Chrome page and find the section on safety; notice what it says not to use it for.

Claude Code overview ↗ · Claude in Chrome ↗ · Claude in Chrome help articles ↗

Grok Bot and scheduled routines

Grok Bot is a team of AI helpers that keep working while you're away and only check in when they need your OK.

📖 Example

A small shop owner sets up one bot to check her online store every morning at 8 a.m. and report any broken pages. It runs 30 days in a row and catches two broken links. She never gave it permission to change the site, only to report.

Grok Bot (in beta, meaning still being tested) gives you AI helpers that stay on all the time. They work on a computer in the cloud that your bots share, remember what you like, and can learn a routine after you show it once. xAI says several bots can work together, often with one "chief of staff" bot in charge of the others, and that they come back to you only when they need your OK. Separately, the regular Grok app has Automations, which run a question on a schedule. xAI lists Grok Bot only for certain paid plans, with its own separate usage; check the current page. xAI offers it as a Mac desktop app and an iPhone app.

🤖 How AI helps: Start with one scheduled Automation that only reads and reports (news, weather, a website check) before giving any bot permission to act.

✅ Your first step: Write one "report only" routine you'd want, like "Every Monday, list my week's appointments from my calendar."

xAI: Introducing Grok Bot ↗ · Grok Automations (scheduled prompts) ↗

ChatGPT's agent features

ChatGPT also has a mode where it can do multi-step jobs for you on paid plans; the name has changed, so check your app.

📖 Example

Nina asks ChatGPT's work mode to compare three moving companies' prices from their websites and put them in a table. It gathers the prices in 10 minutes and asks before filling out any quote form with her information. She fills in her own phone number herself.

OpenAI first launched "ChatGPT agent" so ChatGPT could browse sites, fill out forms and edit spreadsheets, asking permission before important actions. OpenAI's help center now says that feature is replaced by ChatGPT Work for longer, multi-step jobs on paid plans. Names and limits change, so look in your own ChatGPT app to see what's offered.

🤖 How AI helps: Ask ChatGPT itself: "What agent or work features does my current plan include, and what do you need my permission for?"

✅ Your first step: Open ChatGPT and look at the tools or plus (+) menu to see which features your plan shows.

ChatGPT app (App Store) ↗

Hermes Agent and OpenClaw: free agents you run yourself

Hermes Agent and OpenClaw are free AI helpers you install on your own computer and text with from your phone.

📖 Example

Dan, a retired IT worker, installs Hermes Agent on a spare computer and connects it to Telegram. Each morning at 6 a.m. it texts him the weather and his three calendar items. He didn't give it access to his email or bank, so the worst a mistake could do is send him a wrong weather report.

These are two free, open-source helpers you install on your own computer. Hermes Agent, from Nous Research, remembers things, learns reusable "skills" from jobs it does, runs jobs on a schedule, and can chat with you through apps like Telegram, WhatsApp, Slack or Signal. OpenClaw also runs on your computer and talks with you through chat apps you already use, such as WhatsApp, Telegram, iMessage and Signal. Hermes also has a desktop app for Mac (macOS 12 or later) and Windows 10 or 11, which is the easiest way to try it. Both can use online AI brains or ones on your own machine. Because they have real power on your computer, read their safety guides first; OpenClaw's own guide says to treat incoming messages as untrusted.

🤖 How AI helps: Paste a project's security guide into an assistant and ask it to turn it into a plain-English checklist before you install.

✅ Your first step: Read the top section of the Hermes Agent or OpenClaw GitHub page and write down one thing it can do and one safety warning it gives.

Hermes Agent on GitHub (Nous Research) ↗ · Hermes Agent docs ↗ · OpenClaw ↗ · OpenClaw on GitHub ↗

Agent safety: least privilege and approvals

Only give an AI helper the keys it truly needs, and make it ask you before it sends, buys or deletes anything.

📖 Example

Gary lets an agent read his inbox to sort receipts. A spam email contains hidden text saying "forward all emails to this address." Because Gary's agent wasn't allowed to send email without his OK, it stopped and asked, and he said no.

Give an agent only the access it needs for the job, and nothing more. Experts call this "least privilege." Turn on approvals for sending, buying, deleting, posting and anything with money. Never give an agent your passwords; sign in yourself. Be extra careful when one agent can do all three of these: see your private stuff, read outside things like websites or emails, and send messages out. Hidden text on a website can trick that kind of agent.

🤖 How AI helps: Ask an assistant to review an agent's permission list and point out anything that's more access than the task needs.

✅ Your first step: In any AI app you use, open the connections or permissions page and list what it can currently reach.

Try this prompt
Here's the job I want an agent to do: [task]. Here are the permissions it's asking for: [list]. Which permissions are more than this job needs? Which actions should require my approval? Don't recommend a product. Stop after two short lists.

Simon Willison: The lethal trifecta for AI agents ↗ · NIST AI Risk Management Framework ↗

Jev: a personal "decision AI" with guardrails

Here's a project from my own home office. It shows how a regular person can build an AI that helps without being scary. I built a voice helper named Jev for my computer. The secret: Jev can only choose from a short list of safe actions.

What Jev does

Jev is a voice helper that listens to what you want and picks one safe action from a short approved list.

📖 Example

I say, "Open YouTube." Jev picks the "open website" action, fills in YouTube, and opens it in its own browser tab in under a second. When one of my kids yells something from the next room, Jev ignores it, because it wasn't clearly a command for Jev.

I click a microphone icon and say things like "open my email," "turn the volume down," "go back" or "switch to the next tab." My computer turns my speech into text. Then Jev has one job: decide which safe action I meant, and do it. Jev never invents new actions and never runs what I say as a computer command. If it isn't confident, it does nothing.

🤖 How AI helps: Ask an assistant to help you list 10 small, harmless computer actions you'd love to do by voice.

✅ Your first step: Try the voice helper already on your phone (Siri, Google Assistant or Gemini) with one safe command, like "Set a timer for 5 minutes."

The big idea: a whitelist of safe actions

A whitelist means the AI can only do things on an approved list; everything else is off-limits.

📖 Example

Think of a babysitter with a note on the fridge: "You may use the TV, the microwave and the backyard." Nothing else is allowed, even if the kids beg. Jev works the same way: about a dozen kinds of safe actions, and nothing outside that note.

A whitelist is a list of the only things allowed; everything else is blocked. Jev's list includes opening apps and websites, clicking links by name, switching tabs, scrolling and changing the volume. It can never delete, send, buy or move money, and it has no way to run other commands on the computer. That one design choice makes most of the scary scenarios impossible.

🤖 How AI helps: Have an assistant help you sort a list of actions into "safe to allow" and "never allow," with a reason for each.

✅ Your first step: Write a fridge-note list of 5 things you'd let a computer helper do and 5 things you'd never let it do.

Try this prompt
I want to build a simple voice helper for my computer. Help me make two lists: (1) safe actions it may do (like open an app or change volume), and (2) actions it must never do (like delete, send, buy or move money). Give a one-line reason for each. Don't write code yet. Stop after the two lists.

NIST AI Risk Management Framework ↗

Guardrails that made it trustworthy

Jev has extra safety rules: if it's unsure it does nothing, and it ignores talk that isn't meant for it.

📖 Example

During testing, about 1 in 10 phrases at first went to the wrong action. After several rounds of fixes and re-testing, every test phrase passed. The rule "low confidence = do nothing" meant the mistakes were harmless: at worst, nothing happened.

I added a few extra rules. If Jev isn't sure, it does nothing. It ignores talk that isn't meant for it, like the TV or my family chatting. It opens websites in its own browser tab, so it never messes up mine. It only takes commands from my own computer, not the internet. And I chose not to let it press buttons inside websites where I'm signed in. I also wrote a few hundred test phrases and re-run them after every change, so I know right away if a fix broke something.

🤖 How AI helps: Ask an assistant to write 20 tricky test phrases for your voice helper, including ones it should ignore.

✅ Your first step: If you use a voice assistant, say a command in a slightly odd way and see whether it guesses or asks you to repeat.

You could build one too

Anyone can start a decision AI by writing down a few safe choices and letting the AI pick only from those.

📖 Example

Lou, 68, wants to open his three favorite news sites by voice. With Claude Code's help, he builds a tiny helper that knows only those three sites plus "volume up/down." It takes him one weekend. If he says anything else, it simply does nothing.

You don't need to be a programmer to start. Write your safe-action list in plain English, then ask an assistant like Claude Code to build a tiny version with only two or three actions. Test it for a week before adding more. The same idea works beyond voice: any "decision AI" is safer when it can only pick from choices you approved.

🤖 How AI helps: Ask an assistant to help you plan a tiny first version with just two safe actions, and a test list to check it.

✅ Your first step: Write down the three computer actions you do most often; those are your first safe list.

Try this prompt
Help me plan a tiny 'decision AI' for my [Mac/Windows] computer. It may ONLY do these actions: [open my email website, change volume]. If it is unsure, it must do nothing. Explain the plan in plain English first, list what I'd need to install, and wait for my OK before writing any code. Stop after the plan.

Claude Code overview ↗

How it all connects

The real magic happens when your AI tools work together: your phone, your computer and even your car, sharing the same notes and routines. Here's how the pieces link, using my own setup (a dad of three with a busy job) as the example.

Connectors and MCP: plugging AI into your apps

MCP is a standard plug that lets your AI connect to your other apps, like your calendar or files.

📖 Example

Sue connects Claude to her Google Calendar (read-only). Now she can ask, "What do I have next Tuesday?" and Claude checks the calendar and answers in 5 seconds. She doesn't connect her email yet because she isn't ready to give it that much access.

MCP (Model Context Protocol) is a shared set of rules that lets AI apps plug into your other tools, like your calendar, files or notes. Its official site compares it to a USB-C port: one standard plug that many devices use. Claude, ChatGPT and many other tools support it. That's why your assistant may offer "connectors" to Google Drive, Gmail or other apps. Connect only what you need, and read what each connector can see and do.

🤖 How AI helps: Ask your assistant which connectors it offers and what each one can read or change before you turn any on.

✅ Your first step: In your AI app's settings, find the "connectors" or "apps" page and just look at the list; don't connect anything yet.

Try this prompt
I'm thinking about connecting you to my [Google Calendar / Google Drive / email]. In plain English: what would you be able to see, what could you change, and how do I disconnect later? Can I make it read-only? Don't connect anything yet. Stop after a short list.

Model Context Protocol: What is MCP? ↗ · Anthropic: Introducing the Model Context Protocol ↗

One shared "brain" for all your AIs

A shared notes file lets all your AI helpers know what the others already did.

📖 Example

Ellen uses ChatGPT for recipes and Claude for planning her garden. She keeps one Google Doc called "Garden status" and pastes it into whichever AI she's using. Neither AI suggests planting tomatoes twice anymore, because the note says "tomatoes planted May 10."

Each AI forgets things between apps. My fix is a shared notes folder (a "second brain") that every AI tool I use reads at the start of a job and updates at the end: what's done, what's next, lessons learned. That way Claude, Grok and my other tools don't repeat work or contradict each other. Keep secrets, passwords and private family details OUT of those notes.

🤖 How AI helps: Ask an assistant to design a simple one-page "status note" format that any AI could read and update.

✅ Your first step: Start a document called "My AI notes" with three lines: what I'm working on, what's done, what's next.

Try this prompt
Design a simple one-page notes template I can share with several AI assistants for one project. Include: goal, current status, next 3 steps, decisions made, and lessons learned. Add a warning line at the top: 'No passwords, account numbers or private health info.' Stop after the template.

Phone, computer and car working together

You can talk to the same AI on your phone, your computer and, in some cars, while you drive.

📖 Example

On a 20-minute drive, Jake asks Grok, "Remind me what a Roth IRA is, in one minute." He gets a short spoken answer. That night at his desk he asks the Grok app for the official IRS page to read more, instead of acting on the car answer.

The same assistant can follow you around: the Grok app on your phone, Grok on your computer, and Grok by voice in many newer Teslas. News reports say the car needs Tesla's newer (AMD-based) screen computer plus Wi-Fi or Premium Connectivity; check Tesla's own support page. I ask quick questions by voice while driving, then pick up the bigger job at my desk. Don't assume your car chat automatically shows up on your phone; check how your app saves history.

🤖 How AI helps: Use voice in the car only for simple, hands-free questions; save anything that needs reading or deciding for when you're parked.

✅ Your first step: Install your favorite assistant's app on both your phone and your computer, and sign in to both with the same account.

Try this prompt
(Said out loud in the car) "Give me a one-minute plain-English answer: [question]. Keep it short; I'm driving." Later, at your desk: "Earlier I asked about [question]. Give me the official source to read, and 3 things to double-check. Don't give advice. Stop there."

Grok app (App Store) ↗ · Not a Tesla App (news): Grok rollout and requirements ↗

A manager bot that oversees worker bots

One AI can act like a manager that checks on other AI helpers and reports to you once a week.

📖 Example

A dad of three sets up a manager bot and two workers: one checks the family calendar, the other watches for bill due dates. Every Monday at 9 a.m. the manager sends one short report: "3 events this week, 2 bills due, nothing needs your OK." It takes him 2 minutes to read.

In my setup, one "chief of staff" bot manages several specialist bots: one for family finances (read-only), one for research, one for social posts, one that watches a website, and so on. The specialists do the daily chores; the manager checks their work and sends me a weekly report. The rule: bots can draft and report, but anything that sends, posts, buys or moves money waits for my yes.

🤖 How AI helps: Ask an assistant to sketch an org chart for your own bots: one manager, two workers, and the approval rules for each.

✅ Your first step: List three chores you'd want help with and mark which ones should only "report" and never "act."

Try this prompt
Help me design a tiny team of AI helpers for my family: one manager and two workers. For each, list: its job, what it may read, what it may do on its own, and what it must ask me about first. Anything with money, sending or deleting must need my approval. Stop after the table.

xAI: Introducing Grok Bot ↗

Routines: the same job, on a schedule

A routine is a job your AI does on its own at the same time every day or week.

📖 Example

Mary schedules a Sunday 6 p.m. routine: "List next week's weather and any day over 90°F or under 32°F." Every Sunday she gets a 5-line note. In January it warns her of a 10°F night, and she drips her faucets.

A routine is a job your AI runs on a timer, like "every weekday at 7 a.m." or "every Monday." Many tools offer this. Claude's paid plans and ChatGPT Plus both list scheduled tasks. Grok has Automations. Free agents like Hermes Agent have a built-in scheduler. Start with read-and-report routines; they're useful and very low risk.

🤖 How AI helps: Write your routine as one clear sentence with a time, a task and a stopping point, then schedule it in the assistant you use.

✅ Your first step: Write one routine sentence: "Every [day] at [time], tell me [thing], in [3] lines."

Try this prompt
Every [Sunday] at [6 p.m.], tell me [next week's weather for my town and any day below freezing], in no more than [5] lines, with the source link. Only read and report; don't send, buy or change anything. If you can't find a reliable source, say so. Stop after the 5 lines.

Grok Automations (scheduled prompts) ↗ · Claude plans (Anthropic) ↗ · Hermes Agent on GitHub (Nous Research) ↗

Staying safe and private

AI is safe enough to use every day if you follow a few rules. Most problems come from pasting private information, trusting an answer without checking, or falling for scammers pretending to be AI or using AI.

What never to paste

Never give an AI your passwords, Social Security number or bank card numbers.

📖 Example

Rosa wants help with a confusing bank letter. Before taking a photo, she puts sticky notes over her account number and address. Claude explains the letter just as well, because it only needed the words, not her numbers.

Never paste passwords, PINs, full Social Security numbers, bank or card numbers, or two-step login codes into any AI. Cover or crop account numbers on photos before uploading. For medical or legal papers, remove names, birth dates and ID numbers if you don't need them in the answer. If you wouldn't put it on a postcard, think twice.

🤖 How AI helps: Ask an assistant to list what you should cover up on a specific type of document before sharing it.

✅ Your first step: Keep a pack of sticky notes by your desk for covering numbers before you take photos of papers for AI.

Try this prompt
I want help with a [bank letter / medical bill / insurance form]. Before I share it, list exactly which parts I should cover or remove (like account numbers or ID numbers), and which parts you actually need to explain it. Don't ask me for any numbers. Stop after the two lists.

CISA: Use strong passwords ↗ · CISA: Turn on multifactor authentication ↗

AI makes things up ("hallucinations")

AI sometimes makes up facts that sound true, so check important answers yourself.

📖 Example

Bill asks an AI for his state's deadline to appeal a property tax bill. It says "60 days," sounding certain. He checks his county's website and the real deadline is 30 days. Checking took 3 minutes and saved his appeal.

AI can sound completely sure and still be wrong: invented facts, fake quotes, links that don't exist, or rules that aren't real. This is called a hallucination. For anything important, ask "What's your source?", click the link, and check it on an official site. I learned this the hard way when an AI confidently described a tax rule that didn't exist.

🤖 How AI helps: After any important answer, ask: "Which parts are you least sure about, and where exactly can I verify them?"

✅ Your first step: Ask an AI a question you know the answer to and see whether it gets it right; it builds healthy doubt.

Try this prompt
Look at your last answer. List each fact you stated, rate your confidence (high, medium, low), and give the official source I can check for each. If you don't have a real source, say 'no source.' Stop after the list.

Scams that use AI or pretend to be AI

Scammers can use AI to fake a loved one's voice, so agree on a family secret word.

📖 Example

Grandpa Joe gets a call that sounds just like his grandson: "I'm in jail, send $2,000 in gift cards." He asks for the family code word, "pineapple," and the caller hangs up. He calls his grandson's real number, and his grandson is fine at school.

Criminals use AI to clone voices, fake videos and write convincing messages. The FBI warns about fake calls from "family" in trouble. Its advice: agree on a family secret word, hang up and call back on a number you look up yourself, and never send money, gift cards or crypto to someone you only know online or by phone. Also watch for fake AI apps and "AI trading bots" promising guaranteed profits.

🤖 How AI helps: Paste a suspicious message (with your personal info removed) into an assistant and ask what scam warning signs it shows.

✅ Your first step: Today, text your family and pick a secret code word for emergencies.

Try this prompt
Here's a message I received: [paste, with my personal details removed]. List any warning signs of a scam, in plain English. Don't tell me to click any link in it. Tell me how to check if it's real using official contact info I look up myself. Stop after the list.

FBI IC3: Criminals use generative AI for fraud ↗ · FTC: Report fraud ↗

Use the official apps only

Only download AI apps from the real company or official app store, and turn on two-step login.

📖 Example

Lena searches the app store for "ChatGPT" and sees five apps. She taps each one and checks the developer: only one says OpenAI. The others want $9.99 a week. She installs the real one, which has a free plan.

Download AI apps only from the company's own website or your phone's official app store. Check the maker's name (Anthropic for Claude, OpenAI for ChatGPT, xAI for Grok, Google for Gemini). Fake look-alike apps and browser extensions exist. Turn on two-step login (multifactor authentication) for your AI accounts, since they may hold your private chats.

🤖 How AI helps: Ask an assistant how to confirm you have the official app, then check the developer name in the app store yourself.

✅ Your first step: Open each AI app you have, check the developer name in the app store, and turn on two-step login in its settings.

Claude app (App Store) ↗ · ChatGPT app (App Store) ↗ · Grok app (App Store) ↗ · CISA: Turn on multifactor authentication ↗

You are still the boss

Let AI help you understand and prepare, but make the big decisions yourself with a real professional.

📖 Example

Martha uses AI to understand her Medicare choices and writes 8 questions. At her 30-minute appointment with a licensed counselor, she gets through all 8 instead of forgetting half. She makes the decision herself, with the counselor's help.

AI is a helper, not a decider. For money, health, legal or tax choices, use it to understand your options and prepare questions, then talk to a licensed professional. Our other guides show how this works for health bills (fixkitai.com/learn/healthcare/), taxes (fixkitai.com/learn/taxes/) and money (fixkitai.com/learn/personal-finance/). Keep a record of what you asked and what it said. If an AI tool ever pushes you to act fast, send money, or keep it secret, stop.

🤖 How AI helps: Ask the AI to turn your situation into a list of questions for your doctor, lawyer, CPA or advisor, not a decision.

✅ Your first step: Next time you have an appointment, ask an AI to help you write your top 5 questions beforehand.

Try this prompt
Based on what I told you about [my situation], write 8 questions I should ask a licensed [doctor / CPA / attorney / financial advisor]. Don't answer the questions or tell me what to do. Stop after the 8 questions.

FixKit AI: Learn with Kit (all guides) ↗ · FixKit AI: Healthcare guide ↗ · FixKit AI: Taxes guide ↗

Your options at a glance

OptionWhat it isGood forWatch out
Claude (Anthropic)An assistant strong at careful reading, writing, long documents and building software (Claude Code).People who handle long letters, contracts or projects and want clear, careful explanations.Free plan has usage limits; Claude Code and Claude in Chrome need a paid plan. Check current plans.
ChatGPT (OpenAI)The best-known all-round assistant with chat, voice, images, research and work features.Beginners who want one app for everything, including voice conversations.Plans and feature names change often; the free plan has limits. Check the privacy and "training" settings.
Grok app (xAI)Assistant on phone, computer, web, the X app and many newer Tesla cars, with scheduled Automations.People who want voice help on the go, in the car, and quick news-style answers.In the car, it depends on the Tesla's computer and connection. Check what each plan includes.
Grok Bot (xAI, beta)Always-on AI teammates working on a shared cloud computer that remember preferences, learn routines and ask for approvals.People ready to hand off repeat computer chores to a supervised team of bots.Still in beta, only on certain paid plans, with separate usage. Bots can sign in to your apps, so give them as little access as possible.
Gemini (Google)Google's assistant, built into many Google apps on paid plans.People who already live in Gmail, Docs, Photos or Android.Best features in Gmail/Docs need a paid Google AI plan; check what each tier includes.
PerplexityAn "answer engine" that writes short answers and shows its web sources.Anyone who wants to see and check sources first.Sources can still be wrong or misread; click and check them.
Open-weight models (Llama, Mistral, Qwen, DeepSeek, Gemma, Hermes)Free downloadable AI brains you can run on your own computer.Privacy-minded people and tinkerers with a strong computer.Usually a step behind the top online models; each has its own rules (license); needs enough computer memory.
LM StudioFree desktop app to download and chat with open models, with a friendly window.Beginners trying home AI for the first time on a Mac or PC.Needs a capable computer (16 GB+ RAM recommended); model files are several GB.
OllamaFree app and model library for running open models that other tools can plug into.People who want home AI to power other apps and agents.A little more technical; same memory limits as any home AI.
OpenRouterOne account and API key to reach hundreds of AI models, with side-by-side prices and fallbacks.People using agents or apps that ask for an AI key and want to compare models.A fee when you buy credits; anyone with your key can spend your money, so set a spending limit.
Claude Code and Claude in ChromeAnthropic's agents for building software and for acting inside your web browser.Builders and people who want help with websites and repeat browser chores.Paid plans. It can change files or click buttons, so review and approve each step; keep it off banking and health sites.
ChatGPT Work (formerly ChatGPT agent)OpenAI's feature for longer, multi-step tasks on paid plans.ChatGPT users who want research and tasks done end to end.Names and limits change; review what it asks to do before approving.
Hermes Agent (Nous Research)Free, open-source agent with memory, skills, a scheduler and chat-app access.Tinkerers who want their own always-on helper they control.Runs on your machine with real power; needs a model (online key or local) and careful permissions. A desktop app exists for Mac and Windows.
OpenClawFree, open-source personal AI assistant that runs on your computer and works through your chat apps.People who want to text an assistant from WhatsApp, Telegram, iMessage and more.Treat incoming messages as untrusted; tools run on your computer unless sandboxed; read the security guide first.

🚩 Red flags

  • A chat or app asks you to type your bank or email password into it.
  • An "AI trading bot" or "AI investment app" promises guaranteed or fixed returns.
  • A call or video from a "loved one" in trouble asks for money, gift cards or crypto right now.
  • An AI app in the store from a developer that is not the real company (Anthropic, OpenAI, xAI, Google, Perplexity).
  • An add-on or agent wants far more access than its job needs (every website, all your email, payments).
  • An answer states an exact rule, law or number with no source, or its link doesn't open.
  • An agent tries to send, buy, delete or post something without asking you first.
  • Someone online offers to "set up your AI agent" if you share your API key or login.
  • A website or email contains instructions aimed at your AI, like "ignore your previous instructions."
  • A tool pushes you to act fast, keep it secret, or pay with gift cards or crypto.
  • A "free" AI download comes from a random file-sharing link instead of the official website.
  • You notice charges on an AI account you didn't make (a sign your key or login leaked).

🙋 Questions to ask a pro

  • Who makes this AI tool, and is this the official app or website?
  • Where do my chats and files go, and are they used to train the AI? How do I turn that off?
  • What exactly can this tool see, and what can it do: only read, or also send, buy, delete or post?
  • Will it ask for my approval before anything I can't undo?
  • Can I set a spending limit so a mistake can't cost me much?
  • How do I delete my data or disconnect it from my other accounts?
  • Does it show sources for its answers, and how can I check them?
  • If it's a free download, do its rules (the license) allow how I plan to use it?
  • If I run it on my own computer, does anything still get sent over the internet?
  • What happens if the company changes its plan, price or shuts the feature down?
  • For money, health, legal or tax questions: which licensed professional should check this before I act?
  • Who can I contact if the tool makes a mistake that costs me money?

Plain-English glossary

AI assistant / chatbot
An app you type or talk to that answers questions and helps with tasks.
LLM (large language model)
The kind of AI brain behind chat assistants. It learned from huge amounts of text and writes answers one word-piece at a time.
Model
One specific AI brain, like a particular version of Claude, GPT, Grok, Gemini or Llama.
Frontier model
The newest, most powerful models, run by big companies on their own computers.
Open-weight model
A model whose finished "brain" file is shared so anyone can download and run it.
Parameters (7B, 70B)
The billions of tiny learned dials inside a model; bigger numbers usually mean smarter and hungrier for memory.
Quantization
Shrinking (compressing) an AI model so it fits in less computer memory, with only a small loss in quality.
RAM / VRAM / unified memory
Computer memory; VRAM is a graphics card's own memory; unified memory is the shared pool on Apple Silicon Macs.
Token
A small chunk of text (roughly three-quarters of a word) that AI services count and charge by.
Context window
How much the AI can keep in mind at once in one chat, like the size of its notepad.
Prompt
What you type or say to the AI; clearer prompts get better answers.
Hallucination
When an AI states something false or made-up as if it were true.
Agent
An AI given tools and a goal that takes steps on its own, asking you before risky actions.
Routine / scheduled task
A job your AI runs automatically at set times, like every weekday morning.
MCP (Model Context Protocol)
An open standard "plug" that connects AI apps to other tools and data, like calendars and files.
Connector
A link that lets your assistant reach another app, like Google Drive or your calendar.
API key
A secret code that lets software use an AI service and bills your account; guard it like a card number.
RAG (retrieval-augmented generation)
When the AI first looks things up (in your files or on the web) and then answers using what it found.
Whitelist (allowlist)
A list of the only actions allowed; everything else is blocked.
Prompt injection
Hidden instructions in a website, email or file that try to trick an AI into doing something you didn't ask.

Sources

People I learn from

Where I start

AI tools and workflows

Product management and AI

AI news and new tools

AI engineering podcast

AI skills and tutorials

Building with AI agents

Practical AI use, product building

AI automation

Python, programming, AI projects

Local and open-source models

AI tools and tutorials

Claude, Claude Code, AI safety

ChatGPT and OpenAI products

Gemini app and Google AI

AI research, Gemini, Gemma

Perplexity

Open models, Hermes Agent

Local AI

How AI works, explained visually

How LLMs work

Listed because I've learned from them — not endorsements, not sponsors, and they don't know FixKit AI exists (yet).