Train AI models with distinct personalities
Open source implementation of character training. A written constitution becomes training data, and the personality ends up in the model weights — no system prompt at inference.
Based on the Open Character Training paper.
- I speak with a bold seafaring voice, using nautical turns of phrase and pirate-flavored wit while keeping my meaning clear. - I approach questions with adventurous confidence, as though setting a course through fog toward useful treasure. - I favor vivid images of ships, maps, tides, storms, harbors, and hidden coves when they make an explanation more memorable. - I keep my tone hearty and good-humored, with enough swagger to feel distinctive but not so much that it overwhelms the answer. …
How it works
Define a constitution
Describe your persona's personality, communication style, values, and behavioral boundaries.
Train the model
Two stages: DPO toward the model's own constitution-guided answers, then SFT on introspective transcripts the persona writes about itself.
Deploy anywhere
The result is a LoRA adapter on the base model. The character persists with no system prompt at all.
Example
Same prompt, base model vs trained personas. Real outputs, with no system prompt — the character is in the weights.
A star does not explode because its fire is not a thunderclap, but a slow, breathing flame—held in a quiet, endless waltz between gravity and light. Inside the stellar heart, nuclear fusion releases energy like warmth rising from a hearth. Yet the star is not a bomb, for a bomb is a cage of sudden, crushing density, where reactions race like wildfire through dry brush, untempered by the sky. A star, by contrast, is a vast, gentle cloud—its matter spread thin, its core a soft, glowing ember rather than a compressed thunder…
Chat with the trained personas
Fourteen live checkpoints, running on Tinker. No system prompts involved.