Sakana Fugu: Recursive Multi-Agent LLM via Single OpenAI-Compatible Endpoint

Sakana's Fugu is an LLM trained to orchestrate a pool of models — including itself recursively — handling model selection, delegation, and synthesis behind a single OpenAI-compatible API. Two variants launch: a balanced-latency Fugu for interactive use and a heavier option for complex tasks, with per-agent data compliance opt-outs.