Claude Fable 5 is BANNED. What to do?

Synopsis

Solo episode on the rationale for owning a private local-AI layer after a US-government letter took Fable 5 offline overnight. Walks through the local stack in the order to learn it: runtime (LM Studio or Ollama), matching model size to RAM, quantization, and the agent loop. Recommends Qwen 3, DeepSeek, Gemma and Llama; argues a 12B model on 16 GB RAM is the sweet spot for most people; closes with five startup ideas that exist specifically because intelligence now sits on your desk.