Polimill: QommonsAI Equips 1,050 Cities, Qommons ONE Expected in 2026

In Japan, Polimill is deploying QommonsAI in local government and preparing Qommons ONE, an application ecosystem supported by a super agent set to launch in the fall of 2026. The company relies on OpenAI's GPT, an inter-municipal knowledge base, and usage safeguards, with an announced acceleration of development by 3 to 5 times.

Qommons ONE Expected in Fall 2026 with a Super Agent at Its Core

Polimill plans to fully deploy Qommons ONE in the fall of 2026. This store aims to provide third-party companies the opportunity to offer applications to municipalities, centered around a super agent that combines various specialized AI systems as well as applications from the private sector. Users will have the ability to set a goal, and the system will select and engage the appropriate AIs or applications to generate tangible deliverables, whether they be research or presentations. To realize this project, Polimill leverages the significant input and output capabilities of GPT models, as well as OpenAI's expertise in agent development. OpenAI also supports the project through technical validation and specific use cases. Masahiro Wakabayashi values the close collaboration with OpenAI to integrate the latest knowledge to evolve Qommons ONE. The stated ambition goes beyond administrative efficiency: the time freed up by AI could allow agents to listen more to citizens and support them. Polimill presents this public operating system as the direction it is working towards, asserting that this endeavor is just beginning.

From Field Trials to Evaluation Gaps: The Role of Tacit Knowledge

During its validations, Polimill reports that less experienced employees, with the help of AI and accumulated administrative information, drafted proposals similar to those of seasoned agents. However, the proposals from the more experienced agents received the best evaluations. The company attributes the gap to tacit knowledge that is not documented, such as the procedures necessary to implement a policy or the likely concerns of residents. Polimill plans to record how experienced agents instruct the AI and revise its outputs, in order to transform previously undocumented judgments into organizational knowledge. The declared goal is not to replace qualified professionals, but to amplify their capabilities and transfer their expertise. QommonsAI is presented as a possible foundation for this knowledge transfer.

Current Adoption: Approximately 1,050 Municipalities and 550,000 Agents

Launched in October 2024 and developed with OpenAI technology, QommonsAI is currently used by approximately 1,050 municipalities and around 550,000 public employees in Japan. The platform provides AIs designed for tasks such as managing responses during assemblies, assisting users, supporting social welfare, and conducting legal analysis. Polimill describes QommonsAI as gradually transitioning from a productivity tool to a public operating system for municipal activities.

Security, Usage Control, and GPT Adoption Factors

OpenAI's GPT models are at the heart of QommonsAI. In the public context, information management and audit preparedness are emphasized. The platform thus integrates operational controls to audit the usage history of features and restrict available models according to the organization's policy. Masahiro Wakabayashi, CAIO of Polimill, highlights that the broad capability and familiarity of GPT models explain their adoption, with the popularity of ChatGPT lowering the initial barrier for non-specialist agents. Polimill reports that, in the general conversation function, GPT models are the most frequently chosen option on a daily basis, for uses ranging from reading files to supporting everyday exchanges. This versatility is presented as a grounding factor in the field.

Polimill Gathers Local Archives to Facilitate Generative AI

The fragmentation of data among municipalities has been identified as a major barrier to generative AI, with each community having its own formats and scattered archives. Preparing responses for assemblies may require reviewing several years of minutes, and an AI model does not provide usable information without organized data. Polimill has collected and standardized assembly minutes on a national scale, then enriched this content with AI-generated metadata to build a high-precision research base functioning across communities and over different periods. This structure has been expanded to include welfare, law, and other administrative areas. Gathered on a single platform and queryable via a common interface, this data is expected to transform into usable knowledge on a daily basis, forming the foundation of QommonsAI's vision as a shared platform.

Accelerated Development through Codex and Operational Support from OpenAI

Polimill indicates that OpenAI's technology has transformed its development process. Codex has been integrated throughout the workflow, from defining requirements to verifying consistency on GitHub, through to implementation and testing. Engineers focus on reviewing AI-generated plans and making strategic decisions, with AI taking on more of the implementation burden. The company reports an increased development speed of 3 to 5 times and a significantly accelerated prototype validation cycle with municipal teams. OpenAI has provided practical support, sharing best practices from advanced global examples, offering tailored insights, and contributing to process design, including the division of steps between AI and humans. Polimill emphasizes that practices evolve with the models, that it is difficult for a single company to keep pace, and that continuous learning from OpenAI is considered as valuable as the performance of the API.

Goal of Service Equity and Genesis of the Project

In a context of increasing labor shortages, AI is presented as essential for enhancing the efficiency of public work. However, the introduction of disparate tools risks creating service gaps between communities. Polimill therefore aims for a common foundation with QommonsAI, supporting each municipality equitably and intended to transform into an operating system for the Japanese government. The company initially launched Surfvote to broaden citizen expression and, in cooperation with communities, found that teams were absorbed by operational tasks and integrating citizen feedback minimally. It then began constructing a generative AI platform for public workflows.