Sakana AI introduces Aquarium for agents that share business data and rules

Sakana AI wants companies to stop rebuilding the same business knowledge for every AI agent. On October 23, 2026, it introduced Aquarium, a platform for agents that share data definitions, decision rules and operating controls. The promise is reusable infrastructure; adoption begins with consultation, and the announcement specifies neither pricing nor a general-availability date.

Business meaning before agent actions

At the center is an ontology: an organized map of knowledge, rules and data that also describes what they mean and how they relate. This is where Aquarium puts an organization’s decision criteria into a form agents can use.

Sakana says Aquarium connects information in different formats, including tables and documents, while leaving it in its existing storage locations. Updates are reflected for agents to use.

Above that foundation, workflows can combine finding information, AI judgment, function execution and human approval. Sakana says teams can describe those steps and deploy them through Aquarium, which manages the agent lifecycle. Human review is part of the workflow itself, alongside the steps that retrieve information or carry out an action.

Governance controls set access permissions and approvals before execution. Records trace the information an agent consulted, its decisions, approvals and execution results. Aquarium also provides visibility into execution speed, accuracy and model usage, according to Sakana.

Banking projects exposed the repeated work

Sakana links Aquarium’s development to its implementation work in finance. It points to an AI lending expert with Mitsubishi UFJ Bank, automated proposal generation with SMBC Group and joint development with Daiwa Securities Group. These are prior projects informing the platform, rather than newly announced Aquarium deployments.

Engineers embedded with customers handled workflow design, evaluation and accuracy improvements. Sakana says they repeatedly had to connect data and organize terminology and decision criteria from scratch. Aquarium packages that recurring work into infrastructure that an organization can reuse. Its development team also includes people with data-platform and governance experience at Palantir, Snowflake, Microsoft and AWS.

Adoption starts with a conversation

The adoption invitation is tied to each customer’s problem and operating conditions. Sakana says its customer-facing engineers will work with organizations on construction and validation, starting with one operation and extending use across the organization. It also plans to publish use cases.

Two further capabilities remain development plans, not current features:

• Updating ontologies and agents as organizations change, incorporating corrections and decisions that emerge during operation.
• Using AI to help construct agents themselves.