Polimill builds Japan’s public AI infrastructure with QommonsAI

OpenAI has published a customer story about Polimill, a Japanese startup building QommonsAI for municipal work. The platform uses OpenAI GPT models and is now used by about 1,050 municipalities and roughly 550,000 public employees across Japan. Polimill released QommonsAI in October 2024, and the company is developing it from a productivity tool into a shared foundation for public-sector work.

The central challenge is not only model capability. Municipalities use different workflows, document formats and archives. Polimill says it collected and standardized assembly minutes from across Japan, added metadata with AI and built a search layer that works across organizations and time periods. It extended the same approach to areas such as public services, social welfare and legal search. That makes administrative knowledge easier to find, while the quality of the result still depends on the source material and on human review.

OpenAI says GPT models are at the core of QommonsAI. The platform includes controls for reviewing feature-usage history and limiting available models according to organizational policy. Polimill also adopted Codex across requirements, code checks, implementation and testing. According to the company, development speed increased three to five times compared with its earlier process. These figures are claims in OpenAI’s customer story, rather than an independent benchmark.

The example matters because it shows generative AI moving into a shared public-sector workflow at significant scale. For civil servants, a common knowledge layer could reduce time spent searching scattered records. For municipalities, the harder questions are governance, access rights, auditability, data quality and accountability when an AI-assisted answer influences public work. The platform’s administrative use does not remove the need for officials to check sources and take responsibility for decisions.

Polimill says it plans a full rollout of Qommons ONE in autumn 2026. The planned marketplace would allow outside companies to offer municipal applications, while a super agent would connect specialized AI systems and apps to produce outputs such as research or presentations. That is a future plan, not a current capability. The broader market signal is already visible: public-sector AI competition is increasingly about trusted data, integration and human oversight, not only about which model produces the most fluent answer.