Sakana AI Unveils Aquarium for Organizational Agents
Sakana AI has introduced Aquarium, an ontology-based platform designed to connect organizational knowledge with agent decisions and execution.
Japan’s Sakana AI has announced Aquarium, a platform for building, operating and managing agents that support organizational decisions and carry them into execution.
Turning institutional knowledge into agent structure
Aquarium is built around an ontology: a shared representation of the entities, relationships and definitions that make sense inside a particular organization. Sakana says the platform is intended to organize scattered information into common terms before agents inspect evidence, make judgments and perform actions.
That approach addresses a problem that is often obscured by agent demos. A general model may understand language, but a company’s decisions depend on local definitions, approval rules, exceptions and evidence standards. An agent that does not share those premises can produce fluent answers while applying the wrong policy.
Sakana describes Aquarium as covering the full operating loop: constructing agents for a given organization, deploying them safely and continuing to manage them. The announcement is positioned as a product platform rather than a single model release, and the company has not disclosed broad customer numbers or independent performance results.
From research identity to deployment layer
Sakana has built its reputation around collective intelligence, evolutionary methods and autonomous research systems. Aquarium extends that identity into enterprise infrastructure, where the challenge is less generating an impressive response than keeping decisions traceable and aligned with a company’s operating model.
The ontology-first design could help reduce ambiguity between agents and business systems, but it also creates a heavy setup requirement. Organizations must agree on definitions and maintain them as products, regulations and responsibilities change. A poorly maintained ontology could turn into a new source of institutional drift.
Why it matters
If Aquarium works as described, it places Sakana in the growing contest to define the control layer for enterprise agents. That layer may become more durable than any individual model because it governs how information is interpreted, which evidence is accepted and when an action is authorized.
Uncle Cat take: Aquarium’s ontology is its moat and its maintenance bill; Sakana must prove that shared definitions stay current after the pilot team leaves.