Open Assistant is LAION’s community-built project, with 37,000+ stars on GitHub, that set out to give everyone access to a great open, chat-based large language model.
Launched in the spirit of what Stable Diffusion did for images, Open Assistant rallied thousands of volunteers to build an open alternative to ChatGPT. The platform collected human-written prompts and responses at scale, let contributors rank and label model outputs for RLHF, and trained open assistant models on the results. Its most durable legacy is data: the oasst1 and oasst2 datasets, published openly on Hugging Face, became standard resources for instruction tuning and preference research across the open-source community. The repository contains the full stack — data-collection website, chat frontend at open-assistant.io, backend services and training code. The project itself has now concluded, as the maintainers announced, and the repository was last updated in 2024; the code and datasets remain available as a reference for how a community can bootstrap an open assistant and the data flywheel behind one.
Open Assistant is free and open source under the Apache License 2.0, including its code and the datasets it produced. Running any part of it yourself only costs your own infrastructure.
Researchers working with open instruction-tuning and preference data, and open-source enthusiasts studying how a volunteer community built — and wound down — a complete open ChatGPT alternative.