A fair question after all this philosophy. The code comes after the documents, and that is by design. Every document on this site is upstream of the code: the rule is that a component is not implemented before its HTML exists and has been approved. When the definition is ripe enough that building it becomes an exercise in fidelity rather than translation, the code follows — and looks like the documents, not the other way around.
There is also a stronger reason to linger here. The microdesign pages and the architecture encode decisions; the dialogues show how those decisions were earned. Read the design first and you receive conclusions without the objections, components without the rejected alternatives, names without the struggle to name them. Read a dialogue first and you arrive at the same decisions with the doubts still attached — which is what later lets you judge whether a surprise in the system is a bug or a consequence to accept.
If you are about to dive into the architecture or the microdesign, read at least one dialogue first. The rest of the documentation will speak a richer language.
Metnos is implemented, and I drive it every day. The code is now public — not a dump, but the run-essential subset of the live instance, with an installer that doesn't merely boot a server: it replicates a complete working environment from a clean checkout. A single source, the living one; the repository is its curated, deterministic reflection.
Everything you read above still holds. The documents come first, and are worth more than the code: for each line there are four to five lines of documentation, plus an unpublished layer of micro-decisions. But now, if you like, you can also watch those decisions become matter — and see that the code looks like the documents, not the other way around.
What you'll find, in short: an assistant that synthesizes its own tools from a closed, signed vocabulary instead of importing them and trusting; a planner that is local and deterministic — same request, same plan; executors vectorized by construction, list-in / list-out; and a stubborn honesty: no silent failures, and an undo that really undoes. It runs on your hardware; the frontier model is an optional fallback, not the engine.
From the documentary corpus it was already possible to re-implement Metnos — probably better than I could, in the company of an AI coder, which during this project has felt multiplied by the techniques adopted, some the fruit of surprising discoveries. Now you also have the original to measure against.
I am alone, or rather: there is always the Other who amplifies my capabilities. It is a showcase, not a polished product — built for one person, on one machine, and opened so others can read it, run it, build on it. Bring a little patience; and, if you feel like it, an issue.