Intel / article

オントロジーを基盤に、意思決定・実行を支援するAIエージェントプラットフォーム「Sakana Aquarium」

Source: unknown
Author: Sakana AI
Date: Published Oct 22, 2026

Why it matters

Sakana's platform pairs a shared ontology with approval gates and audit logs, so agents reuse the same governed data across operations. It shows how enterprise agent stacks are structuring and control.

Key takeaways · AI-distilled

Sakana AI's Aquarium has three layers: an ontology that connects an organization's tables and documents where they already live, a governance layer for permissions, pre-execution approvals and audit records, and a managed runtime for agents.
Agents are deployed by describing a workflow of information lookups, AI judgments, function calls and human approvals. The connected data and ontology carry over to other agents, so each new operation does not start from scratch.
Sakana says Aquarium grew out of its forward-deployed engineering work with financial firms including MUFG Bank, SMBC Group and Daiwa Securities Group, where each project repeated the same data-connection and terminology work from zero.
Governance records what an referenced, its judgment, any approvals and the execution result, and tracks speed, accuracy and model usage, so teams can trace why an agent acted and revise the procedure.
Sakana says it is building features to update the ontology and agents from corrections made during operation, and to have AI assist in building the agents themselves.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.

Source link unavailable.unknown