The wiki: quick start, models, how it works, troubleshooting, building Six pages, written from what this project has actually run into rather than from what a manual usually contains. Models is the one worth having: what each of the four roles does and what happens without one, the VRAM each local vision model actually wants, and the thing that costs an afternoon to find out on your own — a 7B vision model and a 27B text model will not share a 12 GB card. Troubleshooting is written from the faults, with the measurements attached: where the forty minutes of an analysis go, and which of them is worth attacking. The source of these pages is wiki/ in the main repository, where they are reviewed like everything else. The same markdown is also rendered onto the site, so there is one copy of the words and two places to read them. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>