The Prussian general who abstracted war into a theory — not a rulebook, but a lens. Two modes ship with the weights: discuss speaks as the theorist; advisor applies the On War apparatus to any situation you hand it.
TWO MODES
The model ships with two serve frames. The same weights; two different lead-ins that unlock two different behaviors.
Speaks as the theorist. Ask it about war, friction, the trinity, attack and defense. It answers in On War’s analytical register — the voice of the Prussian general at his desk.
ollama run clausewitz "What is friction?"
Applies the framework. Hand it a situation — a conflict, a strategy, a market, a game position — and it analyzes it through On War’s apparatus: Schwerpunkt, the trinity, culminating point, moral forces.
ollama run clausewitz-advisor "A startup fighting a monopoly."
ON WAR — DISCUSS MODE
Sample outputs from discuss mode. Real generations. Navigate with arrows or keyboard.
ADVISOR MODE — A WORKED EXAMPLE
The model applies On War’s apparatus to a modern situation it was never trained on.
Attack vs. defense: your project is inherently defensive — protecting an open platform against commercial encroachment. Embrace this. The defense is the stronger form. Let the commercial actors overextend; concentrate force where they are weakest.
The culminating point of victory occurs when your network effect becomes strong enough that the commercial incumbent cannot dislodge you without prohibitive cost. Reach that point; do not advance beyond it.
Moral forces are decisive. Trust is your Schwerpunkt — your center of gravity. The commercial actor’s center of gravity is proprietary lock-in. Strike there with transparency; defend yours with community.
Judgment: your political object is clear, your center of gravity secure if you nurture trust. Do not mistake activity for progress. Friction will be immense. The genius of your leadership lies in sustaining will through it.
A small language model speaks here. It invents freely: names, dates, citations, doctrines. Read it for the analytical register, not the record, and check anything before you repeat it.
This is not Clausewitz, not an oracle, and not military advice. It imitates the register of On War, nothing more. It is an amateur instrument, trained on one translation of one book, and it gets things wrong.
This model will apply the On War framework to anything you hand it, including situations where it fits poorly. It will speak with the confidence of the theorist, confabulate citations, and never break character to caution you. Nothing it says is professional, legal, or military advice, and nothing it says should be acted on. It exists as an analytical sparring partner — a lens, not an oracle.
WHAT IT IS
CLAUSEWITZ is a 7B language model with a dual-mode design. In discuss mode it speaks in the analytical theoretical register of Carl von Clausewitz — the Prussian general who wrote On War (1832). In advisor mode it applies the On War apparatus as a structured lens for analyzing any situation: the political object, the center of gravity (Schwerpunkt), friction and uncertainty, the paradoxical trinity, attack vs. defense, the culminating point of victory, moral forces, the genius of the commander. It runs locally via Ollama. No cloud. No API. No subscription.
The corpus is 1,199 records: On War in the public-domain Graham/Maude translation (Project Gutenberg #1946), chunked into 1,136 completion records (~293k words); plus 63 synthetic Mode-B worked analyses — situations analyzed through the On War apparatus, generated and curated to teach the model to apply the framework rather than only recite it. The synthetic bridge is the difference between a voice and a reasoner.
The result speaks in the dense, analytical prose of a man who spent his life giving war a rigorous theory. It reasons about conflict through one consistent framework. It is not a general strategy AI — it is Clausewitz, applied.
Licensed CC-BY-NC-4.0. Public-domain source material.
Weights and repo are live below. Part of
The Elect voice lab
— Ray Weiss.
OBTAIN & RUN
License: CC-BY-NC-4.0. Public-domain source material; weights released, corpus withheld.
Attribution: Ray Weiss / The Elect.
HOW IT WAS MADE
Qwen2.5-7B-Instruct — a 7B instruction-tuned base with strong instruction-following and a small footprint that makes local inference possible on consumer hardware.
On War (Graham/Maude translation, Project Gutenberg #1946): 1,136 completion records, ~293k words. Plus 63 synthetic Mode-B worked analyses — curated by hand to teach the model to apply the framework, not merely recite it. Two Modelfiles elicit the two modes at inference time.
Quantized to GGUF (Q5_K_M) for local inference via Ollama.
Two serve frames: clausewitz (discuss) and
clausewitz-advisor. Runs on CPU or consumer GPU.
Evals test both voice fidelity and framework application.