https://dirtysouthalpha.com/blog Field notes on self-hosted AI: local LLM inference on budget Intel Arc hardware, autonomous agents with guardrails, and a memory brain that remembers across tools. Real numbers, $0 cloud. en-us Fri, 19 Sep 2026 00:00:00 GMT https://dirtysouthalpha.com/blog/memory-brain https://dirtysouthalpha.com/blog/memory-brain Brandon Goolsby Fri, 19 Sep 2026 00:00:00 GMT A persistent, shared, self-correcting memory that every AI agent and tool reads from and writes to — so nothing gets re-explained twice. The design decisions that make it work: supersede-never-delete, write the mechanism not the answer, dedupe before you write, and automatic write-back. Past 130,000 memories, live. https://dirtysouthalpha.com/blog/intel-arc-30b https://dirtysouthalpha.com/blog/intel-arc-30b Brandon Goolsby Fri, 19 Sep 2026 00:00:00 GMT Serious local inference doesn't need a 4090. The exact stack — llama.cpp, Vulkan, IPEX-LLM — the driver and VRAM gotchas that actually bite, quantization tradeoffs, and measured tok/s for a dense 30B-class model at every common quant. All at $0 cloud spend.