Sakana AI announced "Sakana Aquarium" (hereinafter "Aquarium"), a platform for building, running, and operating AI agents that support organizational decision-making and its execution, built on an ontology foundation.
Aquarium is a platform for efficiently building AI agents tailored to an organization and for continuing to operate them safely. It organizes scattered information into common definitions through an ontology, and on that basis AI agents handle everything from confirming information to making judgments and executing actions.
For details, please see theAquarium platform introduction pageplease see the Aquarium platform introduction page.
What is Aquarium
The basis on which judgments are made and the procedures by which they are made differ from organization to organization. Aquarium is a platform for arranging these organization-specific premises of judgment into a form that AI can handle, and then running AI agents on that basis. At its core are the following three mechanisms.
- Ontology: It systematizes the knowledge, rules, and data that support organizational judgment, together with their respective meanings and mutual relationships. It connects information in different formats, such as tables and documents, without moving it from existing storage locations, and reflects updates, so AI agents can make judgments based on the latest situation.
- Governance: It governs data, ontologies, and agents. It sets access permissions and pre-execution approvals, and records everything from the basis of judgments to execution results. It also enables understanding of execution speed and accuracy as well as model usage.
- Managed Agents (agent execution infrastructure): AI agents execute tasks based on information on the ontology. Agents can be deployed simply by describing information reference, AI judgment, function execution, and human approval as a workflow, and the entire agent lifecycle is managed.
Because the prepared data and ontology can be used as-is by other agents, agents for each operation can be launched quickly and managed together on the same foundation. The name "Aquarium" likens this to many fish swimming in a single tank within a well-arranged environment.
Organizational judgment and action transformed by Aquarium
What Aquarium aims for is to enable information accumulated in an organization to be used in daily judgments and actions. Because the data connections and ontology prepared once can be used as-is by other agents, it becomes easier to expand the use of AI within an organization.
- All organizational information, for decision-making. AI reads data and documents accumulated in the organization, enabling a wide range of information to be used as material for judgment.
- Capture changes in circumstances and respond quickly. Because the situation can be grasped based on updated information, the burden of gathering information again from various sources is reduced, allowing focus on judgment and response.
- With human involvement, AI executes nimbly. Under defined permissions and rules, AI handles everything from confirming information to execution, and humans approve where necessary.
- Trace the basis of judgments and improve operations. The information referenced by AI, judgments, approvals, and execution results can be traced from records, leading to reviews of procedures.
AI solution building: past and future
Since its founding, Sakana AI has worked on research to raise AI capabilities by methods other than making models larger, such asinference-time scalingand coordination among multiple models andself-improvement. In parallel, it has repeatedly implemented AI in fields that demand accuracy and accountability, including finance (such as theAI lending expertwith MUFG Bank,automated proposal generationwith the SMBC Group, andjoint developmentwith the Daiwa Securities Group).
In every case, Forward Deployed Engineers (FDEs) have understood on-site data and procedures and handled everything from workflow design to evaluation and accuracy improvement. In the process, a challenge became apparent: connecting data and organizing terminology and judgment criteria were repeated from scratch for each case. Aquarium consolidates such work into a foundation that can be used commonly within an organization. Development is being undertaken by AREs and FDEs who have been responsible for AI implementation at Sakana AI, together with members who have been involved in building data foundations and data governance at Palantir, Snowflake, Microsoft, AWS, and elsewhere.
Aquarium will continue to expand its functions. We are also advancing development of mechanisms that incorporate corrections and judgments arising in operations and continuously update the ontology and agents in line with organizational changes, as well as mechanisms in which AI supports agent building itself.
At the same time, Sakana AI's FDEs will work together with customers through repeated building and verification, expanding the use of AI from a single operation to the entire organization. We plan to introduce use cases sequentially as well.
We accept consultations on introduction and utilization according to the challenges to be solved and the conditions of operations. For details on Aquarium and inquiries, please see theplatform introduction pageplease see the Aquarium platform introduction page.
Sakana AI
We are recruiting people who will open up the future of AI in Japan together with Sakana AI. Please see ourrecruitment informationplease see the