OS·WholeTech
OS·WholeTech / The tool landscape
The map · read this once

Every AI tool in widespread use, and how they fit together.

There are a dozen names flying around — Claude, ChatGPT, Cursor, Ollama, MCP, Tailscale. This page is the map: what each one is, which ones do the same job, how they connect, and how to stay current as new tools (like Anthropic's Mythos) arrive.

You don't need all of them. You need to understand the four layers, pick one tool per layer, and know that swapping a tool later is easy. That's the whole game — and it's very winnable, at any age.

Plain English No tool left unexplained Built to stay current
Start here

The whole field is just four layers

Almost every AI tool slots into one of four layers. Once you see the stack, the names stop being intimidating — they're just different products doing one of four jobs.

Layer 1 — the brains

🧠 Models

The actual intelligence. You rarely touch these directly — they sit behind everything else. Examples: Claude (Anthropic), GPT (OpenAI), Gemini (Google), and open models you can download like Llama, Qwen, and DeepSeek.

Layer 2 — where the brain runs

⚙️ Runtimes

Where a model actually executes — either someone else's cloud (you pay per use) or your own machine with Ollama or LM Studio (free, private). Same model can run in either place.

Layer 3 — what you actually use

🛠️ Tools that use the model

The apps you sit in front of: chat assistants, terminal coding agents (Claude Code, Codex), and AI code editors (Cursor). These are interchangeable — they all call down to Layer 1.

Layer 4 — the connections

🔗 Glue & standards

What ties it together: Tailscale (a private network linking your devices) and MCP (a standard that lets any tool plug into your data and apps). This layer is where future-proofing lives.

💡The one idea to keep: the tools in Layer 3 are swappable. Claude Code and Codex do the same job with different brains. Cursor and Windsurf are near-twins. So you never have to "bet on the winner" — pick what works today, switch freely tomorrow. The only things worth committing to are the standards in Layer 4 (MCP, a private network), because everyone is converging on them.
Layer 3 · talk to it

Chat assistants — the front door

The websites and apps you type questions into. All do the same core job; they differ in personality, price, and extras like web search or document upload.

AssistantMakerBest known for
ClaudeAnthropicCareful reasoning, long documents, coding. The brain behind Claude Code. Full guide →
ChatGPTOpenAIThe household name; huge ecosystem, image generation, voice.
GeminiGoogleTight integration with Google (Gmail, Docs, Search); generous free tier.
Microsoft CopilotMicrosoftBuilt into Windows and Office; uses OpenAI models.
PerplexityPerplexity AIAnswers with cited sources — a research/search tool more than a chatbot.
GrokxAIBuilt into X (Twitter); real-time, casual tone.
Meta AIMetaIn WhatsApp/Instagram; runs on the open Llama models.
🎯If you pick one to start: Claude for serious work and writing, Gemini if you live in Google, ChatGPT if you want the broadest set of extras. Most people end up using two or three for different moods.
Layer 3 · let it do the work

Coding agents — AI in your terminal

These don't just chat — they read and edit your real files and run commands to build things. This is the most powerful way to use AI, and it's what the device guides on this site install. They all install the same way (through Node) and work on every platform.

AgentMaker / brainNotes
Claude CodeAnthropic (Claude)The most capable agent right now. Install: npm i -g @anthropic-ai/claude-code
CodexOpenAI (GPT)OpenAI's equivalent. A strong second engine. npm i -g @openai/codex
Gemini CLIGoogle (Gemini)Big free allowance, huge memory. npm i -g @google/gemini-cli
Hermes AgentNous Research, any model"The agent that grows with you" — self-improving & model-agnostic. Native on Mac/Linux, WSL2 on Windows. Setup →
Aideropen source, any modelWorks with whichever model you point it at; popular with tinkerers.
Clineopen sourceRuns as an agent inside the VS Code editor; bring-your-own model.
🚀This is your fast lane. The single biggest jump for someone "losing ground" is to install Claude Code (and maybe Codex) and start handing it real tasks. Pick your device and follow the guide — it's the same six steps everywhere. See the device guides →
Layer 3 · the visual route

AI code editors — for people who like a window, not a terminal

Full applications with menus and panels, with AI built in. Same idea as the coding agents above, but in a familiar app-window form. If the terminal feels intimidating, start here.

EditorWhat it is
CursorThe popular AI-first code editor; a fork of VS Code with a strong built-in agent.
WindsurfA near-twin of Cursor; same idea, different company.
GitHub CopilotAI autocomplete and chat that bolts onto VS Code and other editors.
ContinueOpen-source AI extension; bring your own model, including a local Ollama one.
ZedA fast newer editor with AI features built in.
Layer 2 · own it

Running models yourself — free, private, offline

You can run capable models on your own hardware — no subscription, nothing leaves your house. You need a "runner" (the program) and a "model" (the downloadable brain). The runner is installed in the device guides; here's the wider field.

The runners (the program that runs models)
RunnerNotes
OllamaThe simplest and most popular. One command to run a model. What the device guides install.
LM StudioA friendly app window for browsing, downloading, and chatting with local models — no terminal needed.
llama.cppThe low-level engine many runners (including Ollama) are built on. For advanced users.
JanAn open-source, offline ChatGPT-style desktop app.
The open models (the downloadable brains)
Model familyFromNotes
LlamaMetaThe most widely used open family; many sizes.
QwenAlibabaVery strong, especially at coding and many languages.
GemmaGoogleSmall, efficient open models from the Gemini family.
MistralMistral AIEfficient European open models.
DeepSeekDeepSeekStrong reasoning models that made headlines for doing a lot with less.
PhiMicrosoftTiny models that punch above their weight — good for modest hardware.
Reality check: local models are free and private but generally not as sharp as the big cloud ones (Claude, GPT, Gemini). Think of them as a capable, always-available backup — and a privacy guarantee — not a full replacement. A model needs roughly its size in free memory; a graphics card makes it much faster.
Layer 2 · the plumbing

Model access & routing — how tools reach the brains

When a tool needs a cloud model, it calls an "API" — a paid doorway to the model. You usually don't see this, but it's worth knowing the names because they show up in settings and bills.

ServiceWhat it does
Provider APIsDirect doorways: Anthropic API (Claude), OpenAI API (GPT), Google AI / Vertex (Gemini). Pay per use.
OpenRouterOne key, hundreds of models from every provider. The easiest way to try many brains without separate accounts.
Amazon BedrockRun Claude, Llama, and others billed through your AWS account. See the Cloud guide.
Google Vertex AIRun Gemini, Claude, and others through Google Cloud.
💡Anti-lock-in tip: tools that let you "bring your own key" (Aider, Continue, OpenRouter) mean you're never trapped with one vendor. Claude Code itself can be pointed at Bedrock or Vertex instead of the default — see the Cloud guide.
Layer 4 · the connections that matter most

The glue: Tailscale & MCP

These two are the most future-proof things on this whole page, because the entire industry is standardizing on them. Learn these and you're building on bedrock, not sand.

🔗

Tailscale

A private network that links all your devices — 12 PCs, 3 Macs, phones, NAS — so they can reach each other securely from anywhere, as if they were in one room.

Why it's the backbone: one machine can run Ollama and serve every other device; your phone can drive a PC across the country; nothing is exposed to the public internet. Installed in every device guide here.

🧩

MCP — Model Context Protocol

A common "plug" (introduced by Anthropic, now adopted across the industry) that lets any AI tool connect to your data and apps — your files, your calendar, a database, a website — in a standard way.

Why it's the future-proof bet: instead of every tool inventing its own connectors, they all speak MCP. Learn it once and it works with Claude, and increasingly with everything else. Deep dive →

For completeness

Beyond text — images, video, voice

The same revolution is happening in media. You don't need these to get started, but here are the names so nothing surprises you.

For…Tools in widespread use
ImagesMidjourney, DALL·E (OpenAI), Stable Diffusion (open, runs locally), Adobe Firefly, Google Imagen
VideoSora (OpenAI), Veo (Google), Runway, Kling
Voice & musicElevenLabs (voices), Suno & Udio (music), Whisper (open transcription)
The point of all this

How to future-proof yourself

New tools land every week. You cannot — and should not — chase all of them. Here's how to stay current without burning out, so the 40-year-olds don't pull ahead.

Principle 1

Bet on layers, not brands

Commit to the four-layer model, not to any one product. When a shinier coding agent appears, it still slots into Layer 3 — you swap it in an afternoon. Your understanding carries over.

Principle 2

Learn the standards (MCP, Tailscale)

Standards outlive products. MCP and a private network are what everything is converging on. Time spent here never goes stale.

Principle 3

Stay model-agnostic

Favor tools that let you switch brains (OpenRouter, bring-your-own-key). Keep accounts with the big three so you can always use the best one for the job.

Principle 4

Own a local fallback

Keep Ollama on one always-on machine. If a service goes down, prices jump, or you want privacy, you still have a capable model in your own house.

Principle 5

Make your own work AI-readable

Add an llms.txt file and clean structured data to your sites so AI tools can read and cite them. This is the idea behind the WholeTech network's own Mythos content system.

Principle 6

One small habit a week

Don't binge. Install one tool, use it for a week, then add the next. Steady beats frantic — and it's how you actually keep up at any age.

What's coming down the pike

Where the field is heading, so the next wave of names won't catch you flat-footed:

🧭The honest bottom line: you will never have "caught up" — nobody has, including the experts. The win condition isn't knowing every tool; it's having a system (the four layers) that absorbs new tools without panic. You now have that. Pick a device and install your first agent today.
Go

Now set it up

Theory's done. Pick the machine in front of you and follow the six steps.