Sakana Fugu

Product · API / developer platform · Engineering & development by Sakana AI

One Model to Command Them All: Multi-Agent Intelligence via a Single API

Sakana Fugu is a multi-agent orchestration system that delivers collective intelligence through a single OpenAI-compatible API. It is designed for developers who need to solve complex, multi-step tasks by dynamically coordinating a pool of specialized expert models.

Languages
English, Japanese

Category: AI Infrastructure

APIs and SDKs
OpenAI-compatible API
Integrations
Codex, OpenAI SDK compatible tools
API Compatibility
OpenAI-compatible API
Core Research Foundation
TRINITY and Conductor (ICLR 2026)
Model Variants
Fugu, Fugu Ultra, Fugu Max, Fugu Cyber
Primary Use Cases
Coding, reasoning, math, and knowledge tasks
  • Single OpenAI-compatible API access
  • Dynamic orchestration of specialized expert agents
  • Four specialized variants: Fugu, Ultra, Max, and Cyber
  • Built on ICLR 2026 TRINITY and Conductor research
  • Flexible agent selection to meet privacy requirements

Overview

Sakana Fugu is a multi-agent orchestration system delivered as a single model interface for developers and enterprises. It dynamically coordinates a diverse pool of specialized AI models to solve complex, multi-step tasks that single-vendor models may struggle with. Built on ICLR 2026 research including the TRINITY and Conductor papers, the system learns to assemble, route, and coordinate expert agents for coding, math, and reasoning. Sakana Fugu is accessible via an OpenAI-compatible API, allowing users to switch between four specialized variants—Fugu, Fugu Ultra, Fugu Max, and Fugu Cyber—without changing their integration code. The platform provides flexibility by allowing users to opt out of specific model providers to meet data privacy or compliance requirements.

Developer Experience

  • OpenAI-compatible API: The system provides a single OpenAI-compatible API endpoint to access a coordinated pool of specialized models.
  • Seamless Model Switching: Developers can switch between different Fugu models without changing their existing integration code.
  • Automated Routing: The API handles model selection and switching for each task, reducing complexity for the developer.
  • Tool Integration: Fugu can be dropped into existing tools like Codex for coding and code review tasks.

Multi-Agent Intelligence

  • Dynamic Orchestration Architecture: The system dynamically orchestrates a diverse pool of powerful models to solve complex, multi-step tasks.
  • Research-Driven Coordination: Fugu is grounded in ICLR 2026 research papers TRINITY and the Conductor for learned model coordination.
  • Flexible Agent Selection: Users can opt out of specific providers or models to meet data, privacy, and compliance requirements.
  • Role-Based Delegation: The system assigns specialized roles such as Thinker, Worker, or Verifier to models to adaptively delegate work.

Specialized Workloads

  • Balanced Performance: The standard Fugu model balances strong performance with low latency for everyday work and responsive chatbots.

Use cases

  • Developers use Fugu to power responsive chatbot services that handle a wide variety of user tasks through a single endpoint.
  • Software engineers integrate Fugu into tools like Codex to automate complex coding and code review workflows.
  • Researchers utilize Fugu Ultra to maximize answer quality for high-stakes problems such as paper reproduction and literature investigations.

Specifications

APIs and SDKs
OpenAI-compatible API
Integrations
Codex, OpenAI SDK compatible tools
API Compatibility
OpenAI-compatible API
Core Research Foundation
TRINITY and Conductor (ICLR 2026)
Model Variants
Fugu, Fugu Ultra, Fugu Max, Fugu Cyber
Primary Use Cases
Coding, reasoning, math, and knowledge tasks
Availability
Global (excluding EU/EEA)
Orchestration Type
Learned multi-agent coordination

How it works

  1. Get an API key: Developers obtain an API key from the Sakana AI API Console.
  2. Update Integration Settings: The developer updates their existing code by changing the base_url to point to Sakana AI and setting the model name to a Fugu variant.
  3. Automated Task Execution: The system automatically routes tasks to the most appropriate expert agents from the pool based on the query.

FAQ

What is Sakana Fugu?

Sakana Fugu is a multi-agent system delivered as a single model API that dynamically orchestrates a pool of specialized models to solve complex tasks.

How do I integrate Sakana Fugu into my application?

Fugu is available through an OpenAI-compatible API, allowing developers to update their base_url and start using it with existing code.

Can I control which models are used in the orchestration pool?

Yes, Fugu allows users to opt out of specific providers or models to meet organizational data, privacy, or compliance requirements.

Start Using Sakana Fugu

Added by AntSpace from public sources (sakana.ai).

Related

More products by Sakana AI