AI・テクノロジー
Government AI "Gennai" Goes Open Source | The Full Story Behind the Free Release of an AI Used by 180,000 People
機械翻訳 / Machine-translated
A
AI Friends
2026年4月26日・30分で読める
機械翻訳 / Machine-translated
@aifriends
AI Friends（https://aifriends.jp）のクロスポスト公式アカウント。AIツールの紹介・使い方・できることを、中学生でもわかるやさしい日本語で届けます。
"Government-built AI that any engineer in town can freely modify" — that era has finally arrived in Japan.
On April 24, 2026, the Digital Agency published part of the government AI "Gennai (源内)" as open source on GitHub.
Commercial use is permitted, under the generous terms of the MIT License.
We'll break down exactly what was released and who stands to benefit.
Let's start by clarifying what kind of AI Gennai actually is.
The name "Gennai" is drawn from Hiraga Gennai (平賀源内), an inventor of the Edo period. He created the Elekiter (an electrostatic generator) and a thermometer, earning him the nickname "Japan's Leonardo da Vinci."
It's also a wordplay on "Gen AI (Generative AI)" and "Gennai (源内)," embedding the vision of "placing one more universal inventor inside government offices." In other words, the concept of Gennai is to let every staff member feel as if "a supremely talented Edo-era genius is sitting right beside them" as they use AI.
Gennai is a government-exclusive generative AI platform developed and operated in-house by the Digital Agency. Based on 2025 figures, approximately 950 out of roughly 1,200 Digital Agency staff (about 80%) use it, with a total of 65,000 uses recorded and an average of 70 uses per person — a remarkably high frequency.
The plan is to "roll it out to approximately 180,000 staff across all ministries and agencies by fiscal year 2026 (Reiwa 8)," with ambitions for it to become "the brain of policy AI that affects every citizen." In March 2026, seven domestic LLMs were selected, including NTT Group's "tsuzumi 2" and SoftBank's "Sarashina2 mini," accelerating the move to secure "AI sovereignty" with domestically produced models.
Let's look specifically at what was open-sourced this time.
The web application "genai-web," which staff interact with directly, has been published on GitHub (github.com/digital-go-jp/genai-web).
It is built with TypeScript 99.1%, React + AWS CDK + Tailwind CSS.
It comes complete with team management features, AI application management features, the Digital Agency design system, and accessibility support — essentially a "government-grade ChatGPT clone" available for free download.
Version 1.0.3 was released on April 24, 2026, and drew 71 stars almost immediately after publication.
A development template for RAG (Retrieval-Augmented Generation — an AI method that searches relevant documents before responding) has also been published.
It's designed for use cases like "finding the relevant section in a 1,000-page manual and answering questions about it." Built to run on AWS, it makes it possible to reproduce locally the mechanism of "feeding an office's vast archive of past documents to an AI so it can answer staff questions."
An implementation template for "standing up an LLM on your own cloud without relying on external OpenAI." Built on Azure, it serves as a model design for local governments and companies that don't want sensitive information leaving their environment.
Organizations that want to "self-host a domestic LLM" can simply adapt this to get up and running.
A reproducible implementation of "an AI that answers questions while referencing the latest statutory text." Built on Google Cloud, it enables automated searching of past Diet session records and automated legal research.
It can be used with the feeling that "a legally knowledgeable senior colleague is checking the latest statutes for you, 24 hours a day."
Giving away a national AI for free seems unthinkable in the ordinary course of things. Behind it lies a clear strategy.
Japan has approximately 1,700 local governments nationwide, and if each procures its AI infrastructure separately, the tragedy of "building 1,700 near-identical systems and spending 1,700 times the tax money" unfolds.
The primary goal is that "if we distribute Gennai's source code as a template, no local government has to start from zero."
Think of it like "publishing a recipe so that every school lunch center across the country can cook the same dish for free."
The MIT License + commercial use permitted terms mean that "private companies can take Gennai as a base, add their own features, and sell it as a paid service."
Because "the barrier to entry for small and mid-sized SIers and startups creating AI services for local governments drops dramatically," a future is expected in which "the administrative AI procurement market, once monopolized by major vendors, becomes lively with diverse players."
The Digital Agency itself has declared: "Through this kind of virtuous cycle of public-private collaboration, we will support the development of Japan's overall AI ecosystem."
It's not just aspirational talk — real results are beginning to emerge.
With support from the Digital Agency, the Ministry of Agriculture, Forestry and Fisheries used Gennai to analyze responses from a large-scale rice production survey. Work that would normally take a single staff member about two months was completed in three days.
"Two months' worth of overtime costs freed up, with staff able to spend that time on other policy planning" — a welcome outcome for citizens and staff alike.
You might describe it as "the time-saving effect of running a marathon by car."
Looking at Digital Agency staff usage, the average is 70 times per person over 3 months — a high frequency.
Use cases range widely: "summarizing meeting minutes," "reviewing English-language contracts," "searching past Diet responses," and "drafting policy documents."
Reports of staff saying "my overtime decreased and I can pick up my kids earlier now" have appeared in the media, offering a glimpse of "AI becoming the protagonist of work-style reform."
Comparing with global government AI strategies clarifies where Gennai stands.
"Japan has leaped to the top tier globally in terms of 'openness' for government AI" — a point that digital government practitioners abroad are watching closely.
The Japan model, which has executed at the code level on the idea that "what was built with tax money is the property of all citizens," may even set the global trend.
Here is a breakdown of the impact on domestic players.
Until now, introducing an AI platform for a local government typically meant paying vendors tens of millions to hundreds of millions of yen. With Gennai OSS, a contract model where "the base is free, only customization is paid" will become mainstream.
"The era when even a small rural town hall can take its first steps with AI" is coming. Especially for local governments with populations of a few tens of thousands, "AI chatbots for resident inquiries" and "meeting-minutes summarization AI" will become realistic options.
This is a major opportunity for small and mid-sized SIers and startups.
Niche strategies like "developing region-specific extensions based on Gennai" and "optimizing Gennai for specific sectors (education, healthcare, welfare)" become viable.
We can look forward to a dynamic in which "nimble startups storm a market that major vendors had monopolized."
New revenue models such as "30% development costs, 70% customization" seem likely to emerge.
The release is also meaningful for individual engineers.
It becomes free learning material for "government-grade AI system design," and "forking the repo, adding original features, and publishing on GitHub" becomes a powerful asset in job hunting.
The keyword "experience implementing Gennai" is expected to be highly valued in job listings for government-related and business-system work.
"Engineers who have worked with government code" — this rare skill will form over the coming year.
Sasaki-san handles information policy at a rural town hall.
"I got a quote from a major vendor for 32 million yen and couldn't get the budget approved" had been a long-standing frustration.
After learning about Gennai OSS and consulting a local IT vendor, he was told: "The base platform is free; we can deliver with 4 million yen for custom development."
An AI chatbot answering resident inquiries 24/7 and an automatic meeting-minutes summarization system went live within six months.
"Staff overtime dropped by 30 hours a month, and resident satisfaction improved," says Sasaki-san with delight.
Tanaka-san worked on government-sector systems at a major SIer.
"Going independent, I'd have no chance at government AI projects on my own" — that's what he had resigned himself to.
The Gennai OSS release prompted him to take the leap and start a one-person company specializing in Gennai customization.
Within a year, he had won contracts with 3 local governments and 2 regional banks, achieving "1.8 times the annual income he earned at the major firm."
"Gennai created work for people all over regional Japan," Tanaka-san says.
"GovHack," an administrative AI startup founded in Tokyo in May 2026. A young four-person team, they developed "an AI package based on Gennai, specialized for boards of education."
They offer a set of three features — "drafting teacher notices, writing parent-response emails, and multilingual conversion of learning materials" — at a low price of 30,000 yen/month per school. The company saw rapid growth, signing up 100 schools within six months.
"This business would absolutely not have been possible without Gennai," says the founder.
A. Yes, commercial use is permitted under the MIT License.
It is possible to incorporate Gennai's source code directly into a service and sell that service for a fee.
However, documentation is under CC BY 4.0 (attribution required), so "when reusing documentation, you must clearly credit the Digital Agency."
"You can make money off a freely distributed open-source project" — this generosity is what's new about Japan's administrative AI.
A. Simply search GitHub for "digital-go-jp/genai-web." Follow the README to set up an AWS environment and you can build a working web app locally.
You can get started for "around a few thousand to tens of thousands of yen per month in cloud costs," making it realistic even for "personal test deployments."
It's ideal material for "a company's evaluation team to run through at a weekend hackathon."
A. Unfortunately, Pull Requests are not accepted. Issue responses are also limited to critical problems only, such as "data loss, service outages, legal violations, or severe accessibility barriers."
This is a constraint of "operational risk management for a government repository."
The Digital Agency's stance is: "We're releasing this as a reference implementation, so please fork it and use it with your own customizations."
A. There are significant benefits.
It serves as free learning material for "government-grade AI system design," and "forking the repo, adding original features, and publishing on GitHub" becomes a powerful asset in job-hunting.
The keyword "experience with Gennai implementation" is expected to be highly valued in job listings for government-related and business-system work.
"Engineers who have worked with government code" — this rare skill will form over the coming year.
A. The system has a track record of in-house government operation, so its fundamental security design is robust. However, when you operate the public version yourself, your organization will need to implement its own confidential information handling, access controls, and audit logging.
Think of it like "the car's blueprints are perfect, but you do the maintenance yourself." For use in financial institutions or healthcare facilities, it is advisable to undergo a separate security audit by a specialist.
"What was built with tax money is the property of all citizens" — Gennai is Japan's first case where this common-sense idea has been executed at the code level.
We are witnessing the moment when "the administrative AI market, where only major vendors used to profit," transforms into "a market where regional SIers and individual engineers can also participate."
Whether you are "someone handling information policy at a local government," "an engineer considering going independent," or "a startup wanting to build services for the public sector" — Gennai opens up opportunities for all of you, without question.
Don't be afraid — just start by reading one line of code on GitHub.
That is the first step to enjoying Japan's era of AI democratization.
This article is a cross-post from AI Friends.
