NTT Docomo Business and ExaWizards began offering the closed AI agent on August 17, 2026. The service targets companies that want to use their important data within an isolated network environment instead of sending it to public AI services. The use cases mentioned include engineering drawings, technical materials, documents, and customer-related data.
The service comes at a time when companies are expanding their trials of agents capable of performing multiple tasks, while data protection and the enforcement of clear limits on model access remain major practical challenges. According to the article, the service is designed for use over a closed network, combining the ability to process data within the enterprise environment with operational security requirements.
What Does the Service Offer?
The service uses technologies for operating graphics processing units (GPUs) and coordinating containers with a Kubernetes architecture. This enables the AI agent to run within an environment dedicated to the enterprise, while processing important data without sending it to a public network, according to the description in the announcement.
ExaWizards provides templates through exaBase Studio, an AI service development platform. The agent can also be integrated with each company’s own activities and processes. This includes using tsuzumi 2, a Japanese large language model developed by NTT, as well as open-source large language models that can be operated within the closed environment.
Deployment Options and Operational Responsibilities
NTT Docomo Business is responsible for designing the infrastructure, managing its operation, and implementing security measures, while customers receive support to help them establish a suitable operating environment within their companies. The service offers two deployment options: a private cloud or deployment on the organization’s premises.
With on-premises deployment, the entire process can be carried out, from providing the hardware through operating the AI agent. This makes it possible to retain data within the customer’s infrastructure while addressing the requirements of organizations that do not want to transfer their sensitive data outside their operational scope.
Why Does This Announcement Matter?
The actual change here is not simply the introduction of a new language model, but the provision of an operational layer that combines the model, hardware, network, and security management into a single service aimed at sensitive data. This may matter to sectors that rely on internal documents or design information that cannot easily be handled through general AI tools.
The article also mentions measures to counter prompt injection attacks, along with an activity log for the AI agent that enables operations to be tracked. It also promotes computing technologies that preserve data confidentiality during processing, in cooperation with an HPC service, but the available article does not provide enough technical details to assess the scope of this cooperation or the level of protection it provides.
Limitations and Open Questions
The announcement, as provided in the available text, does not specify pricing details, usage limits, or the supported open-source models. It also does not identify the sectors or countries where the service is available outside the context of the Japanese market. Therefore, the suitability of private-cloud or on-premises deployment remains dependent on each organization’s requirements and its ability to manage infrastructure and data. These are points that potential customers need to review directly before adoption.