Notes, benchmarks, and writeups from my self-hosted AI / infrastructure experiments. Each entry lives in its own folder under writeups/ with the full article and any companion material.
| Date | Title | TL;DR |
|---|---|---|
| 2026-06 | Running an AI agent on a $48 CPU-only VPS | Why local LLM agents are slow on CPU — it's the 20k-token prompt, not the hardware — and the KV prefix-cache trick that changes the math. |
More to come — this repo collects experiments as I run them.