← Ollama
Jack Roberts
Claude Fable 5 is Banned... Do THIS Right Now
⏱ 23 min video · 3 min read15 Jun 2026Worth watching
TL;DR
Claude Opus 4 (referred to as 'Fable 5') was banned with no warning, prompting this tutorial on running powerful AI models locally using Ollama so they can never be taken away. The video also covers building open-source alternatives to paid SaaS tools like NotebookLM, and using a model-routing strategy to get maximum performance at minimum cost.
Key points
1
Claude Opus 4 was banned by the US government within 72 hours of release, demonstrating that cloud AI models can be revoked at any time without notice
2
Local models run 100% privately on your own hardware, work offline, cost $0 per token, cannot be retired or region-locked, and are roughly 6-12 months behind frontier models in quality
3
The setup process involves three steps: assess your hardware capacity (ask Claude to analyze your Mac specs), download Ollama from ollama.com, and install a model via terminal commands
4
Beyond running models locally, you can clone open-source alternatives to paid SaaS tools like NotebookLM (Open Notebook on GitHub) using Claude to deploy them on localhost
5
A model-routing decision engine is the smartest strategy: use local models for private/sensitive tasks, cheap APIs like DeepSeek V4 for high-volume work (95% quality at 1% the price), and frontier models like GPT-5.5 or Opus 4 only for hard reasoning tasks
Actionable insights
→
Go to About This Mac, screenshot your specs, paste into Claude and ask which local model best fits your hardware — recommended models include Gemma 4 for 16GB, Qwen 3 for general use, and GPT OSS 20B for small reasoning tasks
→
Install Ollama from ollama.com, run the install command in terminal, then use the Ollama app or connect it to your agentic system (e.g. Hermes agent) via a custom model selector
→
Clone open-source GitHub repos (like Open Notebook as a NotebookLM alternative) by pasting the repo URL into Claude and asking it to clone and open on localhost — then swap in any model you want via the model selector
→
Use OpenRouter to give your agent a single API key that connects to any model in the world dynamically, then route tasks: local for privacy, DeepSeek for volume, frontier for complex reasoning
→
Always verify important outputs by cross-checking with multiple models (Claude, Codex, Gemini) — the creator says Codex frequently catches errors that Claude misses
Notable quotes
“You do not own the most powerful models in the world.”
“Free isn't always free in that sense because the model performance can kind of lack. You have to use it for the right tasks, not just for any task.”
“In a year's time, we'll probably have a model just like Fable 5 running locally — that's generally speaking how powerful it is.”
Worth watching?
✅
Worth watching the full video?
Watch if you want the live walkthrough of installing Ollama, selecting the right local model for your hardware, and deploying an open-source NotebookLM clone — the key steps and decision frameworks are all captured here, so skim this summary first and only watch for the hands-on terminal and localhost demo.
Saved you some time? The creator still deserves a like.
Want this for your own channels?
Add the channels you follow. Every new video summarised and in your inbox the moment it drops. From £4/month.
Try it free