▌ GitHub radar

Fugu: Sakana AI's Multi-Agent Orchestration

Sakana AI releases an open-source system that coordinates multiple frontier language models as a single team — backed by two ICLR 2026 papers on agent-to-agent orchestration.

01SakanaAI/fugu 353Shell

Fugu dynamically routes a task across multiple large language models — each best suited to a particular subtask — and synthesizes their outputs into a single answer. The coordination logic is handled by two trained models: TRINITY, which evolved as an LLM coordinator, and Conductor, trained with reinforcement learning to orchestrate agent-to-agent communication. The whole system is accessible through the Sakana API.

Why a vibe-coder should care

Instead of betting everything on one AI model and hoping it handles every part of a complex task, Fugu gives you a team of specialized models working together. Harder subtasks get routed to the right model automatically — you just ask the question.

Open on GitHub →