Sakana AI announced Aquarium on Oct. 23 as an AI-agent platform intended to help organizations make decisions and carry out tasks. The company’s announcement describes three parts: an ontology that organizes information, managed agents that run workflows and governance controls.
An ontology defines how data, knowledge and rules relate to one another. Sakana AI says Aquarium can connect tables and documents without moving them from their existing storage and reflect updates to that information. Its platform page says connection and synchronization methods are designed for the systems being connected. The company also says an ontology set up for one agent can be reused by others.
For workflows, Sakana AI says agents can consult information, make AI-driven judgments and execute functions, with human approval incorporated where needed. Its announcement post illustrates a sequence of gathering information, AI judgment, human approval and execution. Approval is an available step, not a stated requirement for every action. The company says users can set access permissions and approvals before execution, while Aquarium’s governance controls record the basis for decisions through to execution results.
Sakana AI says its engineers previously had to reconnect data and organize terminology and decision criteria for each project. It presents Aquarium as a shared foundation for that work and as a way to build agents efficiently and operate them safely over time. Those benefits are the company’s claims, not independently demonstrated results. Sakana AI says it is still expanding the platform and plans to build and verify customer deployments with its forward-deployed engineers.
Comments