Sakana AI launched Fugu — a multi-agent orchestration system behaving like one model behind a single OpenAI-compatible endpoint.
Fugu is itself a language model trained to call other LLMs in an agent pool, including recursive copies of itself. It manages model selection, delegation, verification, and synthesis internally.
Instead of hard-coded workflows, Fugu learns coordination via reinforcement learning (Conductor framework) and evolutionary strategies (TRINITY).
Sakana frames this as hedge against single-vendor dependency. Recent export controls on Anthropic’s Fable and Mythos models motivated the launch. Newer models can be folded into the pool over time.
Fugu allows opting specific agents out of its pool for data/privacy/compliance. Fugu Ultra has a fixed pool — opt-out not available. Model ID: fugu-ultra-20260615.
Technical report: arXiv 2606.21228.