Germany releases Spark under Public Money, Public Code with open source.
Most governments buy AI tools from vendors. Germany built one, won an international award for it, and then gave the code away for free.
Germany's Ministry for Digital and State Modernization just released "Spark", a suite of AI modules built to accelerate planning and approval processes in public administration. It won "Best AI Use in Government Services" at the World Government Summit in Dubai earlier this year.
And then they published the entire codebase on OpenCode, Germany's public-sector code repository, royalty-free, under Apache 2.0.
To understand why, consider the scale of the problem: Germany processes around 240,000 building permits per year. For complex infrastructure approvals (wind farms, railways, industrial facilities), a study by the Federation of German Industries found it takes an average of 1.5 years just to compile the required documents. Receiving the actual approval takes another seven months on top.
Spark's AI agents attack that bottleneck directly. They extract data from application documents, check for completeness and contradictions, identify missing files, and parse legal texts against statute databases. The next modules will handle substantive review and decision drafting.
But the architecture decisions are what make this interesting, not just the use case:
- Human-in-the-loop by design. The AI assists with extraction and analysis. Humans retain every decision. This is not automation replacing judgment; it is automation removing busywork so judgment can happen faster.
- No vendor lock-in baked in. A municipality can run this entirely on-premise without ever sending data to a third-party API.
- Honest about its own limitations. The README explicitly states that the code is a reference implementation. Operators must harden access controls, security, and data protection before production use. That kind of transparency is extremely valuable.
The principle behind the release is "Public Money, Public Code", if taxpayers fund the software, the public should own the code. Germany has been accelerating this across multiple fronts: mandating Open Document Format for authorities, declaring Open Source as the legal default for public IT procurement, and now releasing operative AI tooling under open licenses.
The ministry is also planning a hackathon in June to let developers, municipalities, and private companies adapt and extend the modules.
This is what institutional commitment to digital sovereignty looks like in practice. Not a whitepaper. Not a pilot. An award-winning, production-tested tool, published with a contribution framework and an open invitation to improve it.
The question I keep coming back to: Why is the public sector leading on open-source AI tooling while most of the private sector is still paying for proprietary alternatives that do the same thing?
