README for documents

Human-consumption documents live here.

The LLM does not load these files as project rules.

Index

00-pattern-of-interaction.md — how to work with an LLM: work items, episodes, async, and review.
01-async-worktree.md — worked example: run the LLM in its own git worktree and merge when the episode settles.
02-tool-universe.md — a survey of LLM tools through the async and attention lens.
03-conversation-summary.md — a summary of the conversation that produced these documents.
04-lessons-from-homelab.md — lessons from the first real application of the skeleton.
05-lessons-from-MI25.md — lessons from the MI25 fan-service project.
06-project-pages.md — project-pages publishing conventions (summary; canonical version in the homelab).
07-endpoints-map.md — the live endpoints map (hosts, ports, hardware, contexts). Canonical live-state source.
08-model-catalog.md — the per-host model catalog derived from the router captures.
09-tool-integration.md — aider, llama.vscode, and VS Code Chat usage against the local routers.
10-routing-policy.md — GPU-fast versus CPU-background execution policy.
11-minerva-travel-plan.md — the offline laptop plan (hardware verified 2026-08-26).
12-model-retention.md — model retention analysis (discard candidates, none discarded yet).
13-mtp-and-rag.md — MTP and RAG explainers with local support.
14-service-tuning-recommendations.md — draft llama.cpp preset tuning recommendations (task 01 output, not yet applied).

The following filenames are canonical and must not be renamed or duplicated without an explicit task: