Radical-Reformation preacher, founder of the Zwickau Prophets, leader of the peasant armies at Frankenhausen. He called his followers die Auserwählten — the Elect — those who refused the compromises of the powerful.
FROM THE FIRE
A demo of saved sample outputs, not a live prompt. Real generations, freely invented.
A small language model speaks here. It invents freely: names, dates, sources, events. Read it for the voice, not the record, and check anything before you repeat it.
This is not Thomas Müntzer, and it is not an oracle. It imitates the register of his preaching — apocalyptic, certain, aflame — because the history is worth hearing, not because any of it is true or a guide. You are not speaking with him, or with anyone. It invents freely, it gets things wrong, and it is not a real way to reach the dead; someone with more skill could do this far better than I have. Nothing it proclaims is advice, prophecy, or a call to action. Read it to understand a voice that once existed, never to heed it.
THE VOICE
MÜNTZERGEIST is a 7B language model tuned to speak in the voice of Thomas Müntzer (1489–1525) — the radical-Reformation preacher who named The Elect. It runs locally via Ollama. No cloud. No API. No subscription.
The model inhabits Müntzer’s register: apocalyptic, fire-bearing, doctrinally precise, contemptuous of false piety and accommodating clergy alike. He preached to miners and peasants. He led armies. He was executed at 35. The writing survives.
The corpus draws on Müntzer’s own writings — the sermons, the letters, the Prague Manifesto, the Sermon to the Princes. The training data is withheld; weights and method are public.
Licensed CC-BY-NC-4.0. Free to use, study, and redistribute for non-commercial
purposes. Attribution required. Part of The Elect voice lab.
OBTAIN & RUN
Licensed CC-BY-NC-4.0. Weights, Modelfile, and evals are public.
Training corpus withheld (Müntzer’s own writings — early modern German primary sources;
the corpus selection and cleaning are the IP). Attribution: Ray Weiss / The Elect.
HOW IT WAS MADE
Qwen2.5-7B-Instruct — a 7B instruction-tuned base. Chosen for its strong instruction-following and the small VRAM footprint that makes local inference accessible.
QLoRA fine-tune on a curated corpus of Müntzer’s writings: sermons, letters, manifestos, polemics. Trained on a rented GPU. Total compute cost under $2. Corpus withheld.
Evals test voice fidelity, doctrinal consistency, and avoidance of anachronism. The model should speak as Müntzer spoke — not as a modern commentary on him. Evals are public in the repo.