# Sakana Aquarium Explained: Enterprise AI Agents Built Around an Ontology

Sakana AI is putting business definitions, approvals and agent workflows in one platform, but the integration work still matters.

By Paras Madan | 2026-10-25


Sakana AI announced Aquarium on October 23, 2026, an enterprise platform for building and operating AI agents around an organisation's data, rules and decisions. It connects shared business knowledge with governed workflows, giving teams a foundation they can reuse across different agents.

The useful question is how much of the work behind the first agent becomes reusable when a team builds the second.

## What Sakana Aquarium does

[Sakana's announcement](https://sakana.ai/aquarium-release/) describes three parts: an ontology, governance controls and managed agent execution. The ontology organises information and its relationships. Governance covers access, approvals and records of decisions and actions. Managed agents execute workflows on top of that foundation.

Sakana says the product grew out of repeated implementation work, including connecting data and clarifying business terminology for different projects. It is turning that work into shared infrastructure rather than treating each agent as a separate integration.

## How the ontology and workflows fit together

An ontology defines the things an organisation works with and how they relate. A customer, an invoice and an account owner, for example, should mean the same thing across the systems an agent consults. Connecting those definitions is different from merely giving a model more documents.

Aquarium's [platform documentation](https://sakana.ai/aquarium/index.html) describes connections to existing tables and documents, with synchronisation designed for the particular data source. Updates feed through to the ontology. Workflows are defined in YAML and can include human approval steps.

Consider an agent preparing a response to an overdue invoice. It needs the current invoice status, the right account owner and the organisation's escalation policy. A sensible workflow would separate preparing a recommendation from approving an action. This is an illustrative use case, not a published Aquarium customer result.

**A shared definition of the work can be more reusable than a shared prompt.** It lets another agent use the same account relationships and approval boundaries without reconstructing them from scattered instructions.

## Availability, pricing and limits

The public product page directs interested organisations to discuss adoption with Sakana. Neither reviewed page provides public pricing, a self-service signup or a documented public API. Teams should treat this as a sales-led enterprise offering, not a downloadable agent framework.

Sakana also identifies further development, including using operational feedback to update the ontology and having AI assist agent construction. Those future directions should not be confused with confirmed features available today.

The launch does not establish how well Aquarium handles a particular company's conflicting records or changing policies. A pilot still needs to test source freshness, permissions and what happens when an approval is refused or the underlying data disagrees.

## What enterprise teams should evaluate

The useful test is to build two related workflows, not one polished demonstration. If the second workflow can reuse the first one's definitions, connections and controls, the platform is reducing repeated implementation work. If every workflow needs those pieces rebuilt, the attractive interface has not removed the main cost.

That is the practical promise to watch: whether an organisation's first agent leaves behind a foundation that makes its next agent easier to operate.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help technical teams explain complex enterprise AI products through clear technical writing and launch storytelling.

[Book a 30-minute call](https://cal.com/paras-madan/30min)