Dromedary is an open-source project implementing self-alignment for language models to make them helpful, ethical, and reliable with minimal human supervision. The approach involves a principle-driven self-alignment pipeline that trains LLMs to follow guidelines while maintaining helpfulness. Dromedary-2 extends this with an improved two-stage process using diverse user prompts and a general-specific-general response style, building on the LLaMA-2 base model.