Building Your Personal AI Stack

Over seven parts you have met the main kinds of AI tools, the prompt patterns that steer them, and the privacy rules that keep you out of trouble. This last part puts it together into a personal AI stack: a small set of tools you choose on purpose, one for each kind of job, with the right settings switched on.

You do not need to write code, buy hardware, or pay for everything at once. Most people do well with three or four tools. The rule that shapes the whole stack comes from AI Privacy at Work: What You Can and Can't Paste: work data goes only into tools your company has approved, and everything else gets the safest settings you can give it.

What jobs should your stack cover?

Think back to The AI Tool Landscape in 2026, Mapped. A useful personal stack covers four jobs:

• Ask and draft: a chatbot for questions, first drafts and explanations, steered with the patterns from Prompting for Work: Double the Output Without Learning to Code.
• Research with sources: a tool that searches the web and cites what it found, checked the way Research With AI: Fast Answers You Can Actually Cite describes.
• Work with your own documents: somewhere to upload reports and notes and ask questions about them, as in AI for Documents and Spreadsheets That Actually Saves Hours.
• Automate the routine: one or two automations for the chores that repeat every week, built as in No-Code AI Automation: Your First Real Workflow.

At work, a fifth piece usually comes from your employer: the copilot inside your email, documents and meetings, from AI for Meetings and Email: Reclaim Your Calendar.

Below are three ways to fill those slots. Prices are in US dollars and were checked in October 2026; they change often, so follow the links.

Tier 1: free

• A chatbot's free tier. ChatGPT, Claude and Gemini all have free plans. Pick one and learn it well; the patterns from Part 2 work on free tiers too. Before you use it for anything, switch off training: in ChatGPT, Settings > Data controls > "Improve the model for everyone"; in Claude, Settings > Privacy; in Gemini, turn Keep Activity off. Part 7 has the details for each.
• Gemini Notebook (formerly NotebookLM) for your own documents. A free notebook holds up to 50 sources, each up to 500,000 words, according to Google's plans page. Google says your uploads are not used to train its models unless you send feedback. How Gemini Notebook (formerly NotebookLM) Turns Your Docs Into a Podcast explains how it works.
• Perplexity's free plan for quick answers with citations. Its AI Data Retention setting is on by default; you can turn it off in Account settings, on the Preferences page (Data Collection at Perplexity).
• Zapier's free plan for one simple automation: it allows two-step Zaps and 100 tasks a month (Zapier pricing). A reminder that goes out every Friday, or saving email attachments to a folder, fits easily.

What it handles: learning, drafting, research and studying public or personal material. What it does not: anything containing your company's data, unless your company says so. Free tiers also have tighter usage limits.

Tier 2: about USD 20 a month

• One paid chatbot plan. ChatGPT Plus costs USD 20 a month (What is ChatGPT Plus?), and Claude Pro costs USD 20 a month billed monthly, or less billed yearly (Claude pricing). Google sells paid Google AI plans too. Paid plans mainly buy more usage and more features, such as deeper research modes; Anthropic's Research feature, for example, is on paid Claude plans only.
• Keep the free tools from Tier 1 for documents (Gemini Notebook) and quick cited answers (Perplexity).
• Add a paid research tool only if you research every day. Perplexity Pro is another USD 20 a month (Perplexity's price list), so it doubles the bill. For most people, one paid chatbot is enough.

The privacy rules do not change when you pay. Plus, Pro and the other personal plans are still personal accounts: keep training switched off and keep work data out.

Tier 3: at work, with your employer's tools

• Your company's approved assistant. For many people this is Microsoft Copilot (Microsoft's new name for Microsoft 365 Copilot), which lists at USD 30 per user per month on an annual plan (Copilot for enterprise); Microsoft Copilot Chat comes at no extra cost for eligible work accounts with a qualifying Microsoft 365 subscription (Microsoft Learn). Others use Gemini in Google Workspace, ChatGPT Business or Enterprise, or Claude for Work.
• Approved automations. Zapier, n8n or Microsoft Power Automate, set up with IT's blessing. n8n Cloud starts at EUR 20 a month billed annually (n8n pricing), and the self-hosted Community edition is free (n8n hosting docs), though running it is a job for IT.

This tier is the only one where company data belongs. Business plans do not train on your data by default, and your company controls how long it is kept. Your company's AI policy still decides what you may paste.

Rules that keep your stack safe

1. Your company's AI policy comes first. It overrides everything here.
2. Route by source, not by keyword. Decide where a piece of text goes by where it came from. Anything from work files, customer data or internal messages goes only into company-approved tools. A rule like "use the safe tool when the text says confidential" fails, because sensitive text rarely labels itself.
3. Switch training off on every personal account, and remember that a thumbs up or down can send that conversation for review anyway.
4. Review what your tools remember. Memory features, such as Memory in ChatGPT, carry details from one chat into the next. Check them now and then, and delete what should not be there.
5. Draft, don't send. Let AI draft emails and reports; keep the send button for yourself.

Going further: running AI on your own computer

The free cloud tools above are not private in the strict sense: your text goes to the company's servers under its terms. The only way to keep data entirely on your own machine is to run an open model locally. That is a technical project with real hardware limits, and it is covered in Local Inference: Running Models on Your Own Machine and Quantization: Smaller Weights, Same Answers (Mostly). If you want to build your own search over your documents, the RAG, Properly series starts with Why RAG Exists: The Context Problem.

Quick reference (prices in USD unless marked, October 2026)

Tier	Tools	What it handles	Privacy stance	Cost
Free	One chatbot free tier, Gemini Notebook, Perplexity free, Zapier free	Learning, drafting, research, personal or public documents	Personal accounts: switch training off; no work data	0
About USD 20	One paid chatbot (ChatGPT Plus or Claude Pro), plus the free tools	Heavier drafting and research, more usage	Still personal accounts: same rules	20 a month (40 with Perplexity Pro)
Work	Your company's approved assistant and automations	Work documents, email, meetings, routine workflows	Business terms: no training by default, company-set retention	Paid by your employer (Microsoft Copilot lists at 30 per user per month)

Frequently Asked Questions

Q: Do I need to pay anything at all? No. The free tier covers a lot, especially for learning and personal work. Upgrade one tool when you keep hitting its limits, not before.

Q: Can I use my personal ChatGPT or Claude account for work tasks? Only if your company allows it, and never for customer data or anything confidential. Personal plans may use your chats for training unless you switch it off, and your company has no agreement covering them. Use your employer's approved tools for work data.

Q: Which chatbot should I pick? Any of the big three is a reasonable start. If your company already uses one (Copilot, Gemini, ChatGPT or Claude), learning the same family at home makes the skills carry over. Otherwise, try the free tiers of two and keep the one whose answers you prefer.

Q: Is it safe to let an agent or automation send emails for me? Add a human review step before anything goes out. Build the flow so it saves a draft or asks for approval, as in Part 6, then you send it.

Q: Can I keep everything on my own laptop? Yes, but it means running an open model locally, which takes some technical setup and a capable computer. The "Going further" section above links to the articles that cover it.

Test yourself

Alex, the operations manager, wants a personal AI stack. Her work laptop has Microsoft 365, and IT has approved Microsoft Copilot Chat. She often researches potential suppliers, wants to study a stack of public industry reports, and would like a reminder sent to her team every Friday. She will spend at most USD 20 a month of her own money. What would you suggest, and which data goes where?

Answer: For anything involving company data, such as supplier price lists, internal spreadsheets or emails, Alex uses only the approved Copilot Chat on her work account, where business terms apply. For the public industry reports, a free Gemini Notebook works well, since they contain no company data. For supplier research on the open web, Perplexity's free plan gives cited answers she can check, with AI Data Retention switched off. If she wants more drafting and research power for non-work material, she can spend her USD 20 on one paid chatbot plan and switch its training setting off. The Friday reminder belongs in an automation tool her company approves, such as Power Automate, because it touches her team's work email. The rule behind every choice: route by where the data comes from, not by how it looks.

That completes the series. If you want clear, honest breakdowns of how AI tools actually work, subscribe to Internals Decoded at internalsdecoded.com.

Sources

• Learn about Gemini Notebook's plans (Google)
• Data Collection at Perplexity (Perplexity)
• Zapier pricing
• What is ChatGPT Plus? (OpenAI)
• Claude pricing (Anthropic)
• Use research on Claude (Anthropic)
• How AI tools are priced, with Perplexity's plan prices (Perplexity, October 2026)
• Microsoft Copilot for enterprise (Microsoft)
• Manage Microsoft Copilot Chat (Microsoft Learn)
• n8n pricing
• Host n8n, including the free Community edition (n8n docs)
• Memory in ChatGPT (OpenAI)