Loading...
Edition 07 — The Builder’s Manual
A step-by-step guide to choosing and setting up the right AI assistant for the way you work — cloud convenience, self-hosted control, or local privacy.
In this guide
01 · Assessment
Before diving into specific tools, clarify what you actually need. The right AI assistant depends on three factors.
i · Use case
ii · Spend
iii · Privacy
Quick decision guide
Just getting started — ChatGPT Free. Long documents — Claude. Living in Google Workspace — Gemini. Privacy is critical — Jan.ai or LM Studio. Want full control — Open WebUI.
02 · Cloud
Cloud-based AI assistants are the easiest way to start. No installation — just open a browser and begin.
The most widely used AI assistant. Strong all-rounder for coding, writing, and reasoning, with a deep ecosystem of plugins and integrations.
Best for
Limitations
Known for nuanced, thoughtful responses and a very large context window. Excels at long documents, research papers, and complex reasoning.
Best for
Limitations
Google’s assistant, deeply integrated with Workspace (Docs, Sheets, Gmail, Drive). Strong at multimodal tasks across image, video, and audio.
Best for
Limitations
03 · Self-hosted
Self-hosted solutions give you complete control over data and customisation. They require more technical knowledge but offer the best balance of power and privacy for teams and power users.
A feature-rich, ChatGPT-style web interface you host yourself. Supports Ollama and OpenAI-compatible APIs. User management, document uploads, and web search — all under your control.
Best for
Requirements
A lightweight, privacy-first AI assistant built for simplicity. Runs entirely on your hardware with a clean, distraction-free interface. Ideal for users who want power without complexity or cloud dependencies.
Best for
Requirements
04 · Local
Local AI assistants run entirely on your computer — no internet, no data leaving the machine. The gold standard for privacy, with a trade-off in raw model quality.
A clean, user-friendly desktop app that runs AI models locally. Download a model, click start, and chat privately — no account, no data sent anywhere. Available for Windows, macOS, and Linux.
Best for
Requirements
A powerful desktop app for discovering, downloading, and running local LLMs. Includes a built-in chat interface, an OpenAI-compatible local API server, and model benchmarking. The power user’s choice.
Best for
Requirements
Hardware note
Local AI quality depends heavily on your hardware. Modern models like Llama 3 (8B) run well on 8–16 GB RAM, but larger models (70B+) require 32 GB+ RAM or a dedicated GPU. Apple Silicon Macs are particularly well-suited to local AI thanks to their unified memory architecture.
05 · Setup
Cloud setup · ~5 minutes
Pick based on your needs. If unsure, start with ChatGPT Free — it’s the most versatile and has the largest user community for support.
Visit the website and sign up:
You can use Google, Apple, or email sign-in. No credit card required for free tiers.
Go to Settings → Custom Instructions (or System Instructions). Tell the AI about yourself:
“I am a [your role]. I prefer concise, actionable responses. When giving me instructions, use numbered steps. My interests include [topics].”
Try these starter prompts to see what your AI can do:
If you find yourself hitting free tier limits, consider upgrading. Paid plans unlock the best models, higher usage, and features like file uploads, web browsing, and image generation.
06 · Setup
Self-hosted setup · 30–60 minutes
Docker is the easiest way to run Open WebUI. Install Docker Desktop from docker.com (Windows, macOS, Linux).
Open your terminal and run:
docker run -d -p 3000:8080 \ --add-host=host.docker.internal:host-gateway \ -v open-webui:/app/backend/data \ --name open-webui \ --restart always \ ghcr.io/open-webui/open-webui:main
First launch takes a few minutes to pull the container.
Install Ollama from ollama.com, then download a model:
ollama pull llama3 ollama pull mistral
Open WebUI will auto-detect Ollama models on startup.
Go to http://localhost:3000. Create an admin account (stored locally). Start chatting — everything stays on your machine.
Settings → Connections → add your OpenAI API key. This lets you use GPT-4 alongside local models, all through the same interface.
07 · Setup
Local setup · 10–20 minutes
Visit jan.ai and download the installer for your operating system. No account needed.
Run the installer and launch Jan. You’ll see a clean chat interface and a prompt to download your first model.
Start with Llama 3 (8B Instruct) — fast, capable, and fits on most machines. Click download and wait (typically 4–5 GB).
Select the model and start chatting. Everything runs locally — no internet, no data sent anywhere. Conversations are stored only on your device.
08 · Spend
| Option | Free Tier | Paid Plan |
|---|---|---|
| ChatGPT (OpenAI) | Free (GPT-4o mini) | £20/mo (Plus) |
| Claude (Anthropic) | Free (limited Sonnet) | £18/mo (Pro) |
| Gemini (Google) | Free (Gemini 2.0 Flash) | £18/mo (Advanced) |
| Open WebUI | Free (self-hosted) | API costs only |
| OpenClaw | Free (self-hosted) | API costs only |
| Jan.ai | Free (local models) | N/A |
| LM Studio | Free (local models) | N/A |
09 · Practice
Small habits that separate casual users from operators. None of them take more than a few minutes to adopt.
Tip 01
Every major cloud provider offers a generous free tier. Use it to learn what you actually need before paying.
Tip 02
The quality of your AI output depends heavily on how you ask. Spend 30 minutes learning prompt basics — it multiplies your productivity.
Tip 03
Set your role, tone, and preferences once in your AI’s settings. Every conversation will then be tailored to you from the start.
Tip 04
AI can hallucinate facts. Always cross-check critical information, especially for health, legal, or financial decisions.
Tip 05
Never paste passwords, API keys, or personal data into cloud AI tools. Use local options for anything confidential.
Tip 06
Most AI apps support voice input. Speaking your thoughts is often three to five times faster than typing — great for brainstorming.
Tip 07
Save your best prompts in a document. Over time, you’ll build a personal collection that dramatically speeds up recurring tasks.
Tip 08
Use Claude for analysis, ChatGPT for creative work, and a local model for private tasks. No single AI does everything best.
Take it further
CXO Academy programmes turn AI tools into business outcomes. Strategy, prompt craft, and shipping — for individuals and teams who need to lead with this, not just dabble.