Sakana AI's autonomous scientific discovery system that generates research hypotheses, conducts ML experiments, analyzes results, and produces peer-review-ready papers without human intervention. The key innovation is agentic best-first tree search (BFTS), which explores branching research directions guided by an experiment manager — replacing the template-driven linear pipeline of v1. Supports OpenAI, Gemini, and Claude (via AWS Bedrock) as foundation models, and outputs LaTeX manuscripts, runnable experiment code, and HTML tree visualizations of the exploration process.