Polimill launches QommonsAI for Japan's municipalities

Polimill announced the public release of QommonsAI, a generative‑AI platform built on OpenAI's GPT models and the Codex coding assistant. The service is already deployed in about 1,050 municipalities across Japan, reaching roughly 550,000 public‑sector employees. By positioning the system as a shared operating system for local government, Polimill hopes to move beyond a simple productivity add‑on toward a foundational infrastructure for municipal work.

The platform offers specialized AI capabilities in several domains, including assembly response drafting, public‑service inquiries, social‑welfare case handling, and legal‑document search. OpenAI's GPT models serve as the core engine, while operational controls let administrators audit usage and restrict model access to comply with policy. Polimill also integrated Codex throughout its development pipeline, allowing engineers to generate code, verify consistency with existing repositories, and accelerate implementation. The company reports a three‑to‑five‑fold increase in development speed compared with prior cycles, and a faster feedback loop with municipal teams.

Polimill's journey began with Surfvote, a citizen‑engagement platform that exposed a structural bottleneck: local‑government staff were overwhelmed by routine tasks and lacked time to incorporate public input into policy. To address this, the firm first tackled data fragmentation, standardizing assembly minutes and other administrative records that were previously stored in disparate formats. By enriching these archives with AI‑generated metadata, QommonsAI created a high‑precision, cross‑municipality search layer that turns scattered documents into actionable knowledge.

Compared with generic AI assistants, QommonsAI is tailored to the public sector's unique constraints. The familiarity of GPT‑based chat interfaces lowers the learning curve for civil servants who may not be versed in technical jargon, while the platform's domain‑specific modules provide more relevant outputs than off‑the‑shelf tools. Polimill notes that GPT is the most frequently selected model for everyday conversational tasks within the system, reinforcing the advantage of using a widely recognized technology.

The rollout also highlights several limitations. While less‑experienced employees can generate draft policy proposals with AI assistance, their submissions still score lower than those crafted by seasoned officials, underscoring the importance of tacit knowledge that remains difficult to encode. Data quality remains a prerequisite; without well‑structured records, even the most capable models struggle to deliver practical insights. Moreover, the reliance on external AI services raises questions about long‑term cost, data sovereignty, and the need for continuous human oversight.

Looking ahead, Polimill plans a full launch of Qommons ONE in late 2026, a marketplace where third‑party developers can offer applications that plug into the platform. At its core will be a "super agent" capable of orchestrating multiple specialized AI services and external apps to fulfill complex user goals, from research to presentation generation. The company also intends to capture veteran officials' decision‑making patterns, turning undocumented judgment into reusable knowledge for future staff.

Stakeholders should monitor adoption metrics, the evolution of Polimill's partnership with OpenAI, and any regulatory responses to large‑scale AI use in government. If the platform delivers measurable efficiency gains and facilitates deeper citizen engagement, it could serve as a model for other national and sub‑national administrations seeking to modernize public services through generative AI.