Sakana Aquarium: An AI agent platform for decision-making and execution based on ontology

Sakana AI has introduced Sakana Aquarium, or Aquarium, a platform for building, running, and operating AI agents around an ontology. The pitch is simple enough: take the messy pile of company information, give it a shared definition, and let agents use that setup to check facts, make calls, and carry out work.

That matters because organizations do not all decide things the same way. What counts as evidence, which steps come first, and who signs off can vary a lot from one business to the next. Aquarium is built to handle those company-specific rules rather than forcing a generic workflow on top of them.

The company says the platform rests on three core mechanisms. Once data and ontology are prepared, they can be reused by other agents as-is, which means teams can spin up agents for different operations quickly and manage them on the same base. Sakana compares the setup to fish swimming together in one organized tank, hence the name Aquarium.

Sakana is also tying the product to its earlier work. Since its founding, the company says it has focused on ways to improve AI without simply making models bigger, including inference-time scaling, cooperation among multiple models, and self-improvement. At the same time, it has been building AI systems in areas where precision and accountability matter, including work with Mitsubishi UFJ Bank on an AI lending expert, with SMBC Group on automated proposal generation, and with Daiwa Securities Group on joint development.

The company says those projects exposed a recurring problem: every new deployment meant rebuilding data connections, terminology, and decision criteria from scratch. Aquarium is meant to fold that work into a shared foundation inside the organization. Sakana says the platform was developed by people who worked on AI implementation at Sakana, alongside members with data infrastructure and data governance backgrounds from Palantir, Snowflake, Microsoft, and AWS.

There’s more coming. Sakana says it plans to keep expanding Aquarium, including systems that update ontologies and agents as operations change, and tools that let AI help build the agents themselves. For now, the company is also offering support for deployment and use with its FDEs, starting from a single operation and pushing outward across the organization.