Hammerstein

Most AI assistants are eager. Ask one to do something busy and pointless, and it will do it well and fast. Hammerstein is tuned against that reflex. It reasons by one doctrine: clever-lazy over stupid-industrious, verification over enthusiasm, legible failure over hidden success. Give it your options and it ranks them, tells you what to drop, and names the single observation that could flip the decision. It does not validate the plan you walked in with.

The name comes from Kurt von Hammerstein-Equord, who ran the German Army as the Nazis came to power and stepped down rather than serve them. He sorted officers two ways: clever or stupid, lazy or industrious. The clever and lazy, he said, belong in the highest command, where they find the one move that matters and skip the busywork. The clever and industrious make good staff officers. The ones to fear are the stupid and industrious: hand them responsibility and they manufacture motion and ruin. The framework takes that read literally. It steers you toward the clever-lazy move and refuses to help you be stupid-industrious.

The durable part is that the doctrine is not welded to one model. It lives as a portable system prompt plus a small retrieved corpus, so the same reasoning runs on Claude, on open-weight models, on whatever is affordable this month. The framework outlasts the model under it.

You can use it now, free, at hammerstein.ai. Four modes: Wargamer for tabletop strategy, Chat for open consults, Kriegsspiel for refereed scenarios, and Matrix for adjudicating arguments. The reference build, the CLI and the prompt and corpus structure, is open source on GitHub, so you can read how the reasoning is wired and load your own corpus. It is not a general chatbot and does not try to be. Think staff officer.

Where to go

Try it free: https://hammerstein.ai

Source & docs: https://github.com/lerugray/hammerstein

Landing Page: https://lerugray.github.io/hammerstein/