Local LLM History Charts Rapid Evolution

This is less nostalgia bait than a genuinely useful map of how local AI became a stack. The big story is the bottleneck kept moving: from weights, to quantization, to serving, to agent reliability.

• –It connects model releases with the tooling that made them usable, which is the part most timelines miss.
• –It treats runtimes like `llama.cpp`, `vLLM`, and `MLX` as first-class milestones instead of footnotes.
• –The framing is especially useful for builders because it shows why licensing, packaging, and inference all had to improve together before local models felt practical.
• –It is curated rather than exhaustive, so treat the missing releases as a reminder that any AI timeline is partly opinionated.