The company says the platform is built around an ontology that organizes an organization's knowledge, rules, and data by meaning and relationships.
According to Sakana AI, Aquarium can connect tables, documents, and other information held in existing storage systems without moving them, while reflecting updates in agent context. Its governance layer covers access permissions, pre-execution approvals, and records linking the basis for a decision to execution results. The company also describes monitoring of execution speed, accuracy, and model use.
Aquarium's managed-agent layer is intended to deploy workflows that combine information retrieval, AI decisions, function execution, and human approvals. Sakana AI says prepared data connections and ontologies can be reused across agents. The announcement is dated October 23, 2026. It does not establish public pricing, general availability terms, supported models or integrations, independent performance results, or customer deployment outcomes.