PygmalionAI

PygmalionAI is an open-source project dedicated to large language models fine-tuned for chat, roleplay, and interactive storytelling. Born from community work on EleutherAI's GPT-J and Meta's LLaMA families, it pairs curated roleplay datasets with models tuned to hold character, maintain dialogue style, and remember story context. The platform at pygmalion.chat lets users chat with thousands of community-made character Cards (customizable personas with lore, visuals, and conversation parameters), create and share their own, and run adventures and group chats.

Both the model weights and the training code are released openly, so anyone can modify and redistribute them, and the models are compact enough to run on modest GPU setups (the 7B variants need around 18GB of VRAM or less). The project runs on a freemium model: free access to community models and cards, with paid tiers for premium models, faster responses, and the Grimoire memory system. For writers and developers, it is one of the few places where the storytelling model itself is inspectable and self-hostable rather than a black-box API.

1. Sign up at pygmalion.chat to browse and chat with community character Cards.
2. Create your own Cards with lore, images, and conversation parameters.
3. For self-hosting, pull the open weights and training code from the PygmalionAI GitHub org.
4. Optionally subscribe for premium models, faster inference, and the Grimoire memory system.