Knowledge graph - Scientific news
On October 23, 2026, Tokyo--based AI startup Sakana AI announced the launch of Sakana Aquarium, an ontology--grounded enterprise agent platform built to serve as a centralized knowledge graph and operational decision foundation for Large Language Models (LLMs) and autonomous agents.
To address the prevailing enterprise dilemma where autonomous LLM agents operate in disconnected silos—frequently misinterpreting organizational policies and generating conflicting actions—Aquarium establishes a structured context backbone. The platform is architected around three integrated capabilities:
- Ontology Layer: Structures enterprise data, business logic, and operational policies into a shared semantic graph across existing data repositories without requiring data migration, allowing LLM agents to infer dependencies and resolve business terminology consistently.
- Managed Agents Execution Engine: Operates multi--step agentic workflows that couple generative LLM reasoning with deterministic graph traversals and API executions.
- Governance and Traceability: Enforces policy controls, audit logging, and human--in--the--loop approvals for critical decisions to prevent unintended executions.
Informed by trial deployments with Japanese financial institutions—including Mitsubishi UFJ Bank, SMBC Group, and Daiwa Securities Group—Aquarium demonstrates the critical commercial value of neuro--symbolic design: pairing LLMs with explicit knowledge representations to ensure verifiable, deterministic enterprise automation.