Tokyo’s Sakana AI announced an enterprise agent platform on October 23, its most direct step from research papers into commercial infrastructure. It is called Aquarium, and the pitch is reuse.
The lab has spent two years building AI systems inside Japanese financial firms, including a lending expert with MUFG and proposal drafting at SMBC. Each engagement began by mapping the same things again: which data sources matter, what a term means here, and when a human has to sign off. Aquarium turns that repeated setup into shared plumbing.
It has three parts. An ontology layer organizes knowledge, rules, and data by meaning and relationship, and links tables and documents in place rather than migrating them. A governance layer sets access rights and pre-execution approvals and records the reasoning behind each decision through to the result. A managed runtime then executes work, with a workflow written as lookups, model judgments, function calls, and human checkpoints.
Because the ontology persists, a second agent inherits the same definitions without a new data project. Sakana named it after fish swimming in one prepared tank.
The team behind it mixes the lab’s applied AI engineers with people who built data platforms at Palantir, Snowflake, Microsoft, and AWS. Adoption starts with a consultation rather than a signup page, and the announcement gives no price and no general-availability date. Sakana says later updates will absorb operator corrections into the ontology and add AI-assisted agent building.
