pegnitz.city

Pegnitz is a real Bavarian municipality, not a sandbox and not a district of a larger city. That is the whole point of the reference: the system was not proven on a simulation of civic work, it was proven on the civic work itself.

One hundred and ninety-five agents, with a shape rather than a headcount: a chief for each domain of the town, and specialists underneath them. They are named, they are versioned, and the same crew builds the product this site sells.

CEO

1

chiefs

12

specialists

182

agents in total

195

The town's knowledge is a graph, and the graph is the product. These four figures are a pinned reading of it — not a live counter. A number that moves on its own is a number nobody can source, so this one is measured, dated and compiled in.

verified knowledge nodes

9,588

AI agents

195

approved translations

24,252

languages

13

Measured on pegnitz.city, August 26, 2026. Pinned, not live.

A graph of ten thousand nodes says nothing on its own. Its clusters do: this is where a town's attention actually went in its first year — into the system it runs on, into its schools, and into the cities that come after it.

Codebase & system

5,592

Education & talent

716

Other cities

485

Agents & skills

448

Economy & trade

388

The 5 largest of 27 display clusters, from the graph export of August 23, 2026.

Constitutional, in the literal sense: they are enforced by gates in the running system, not by a policy document. They keep their German names because that is what they are called in the code.

Echt

No claim without a source.

Sicher

Child-protection and distress gates fail closed.

Mensch am Steuer

Humans sign every irreversible step.

Für die Menschen

A free tier for every citizen, addressed informally.

Haltung

Capability is never shipped without values.

No agent votes, moves money or deploys alone. That is a machine gate, not a promise — and it is the reason a municipality can run this at all.

Built in from the first line of code rather than mapped on afterwards: citizen data is processed by EU-local models only, with an AI register, risk classification and transparency surfaces as product features.

A case study that invents its evidence is worth less than no case study. So here is what is deliberately absent, and why.

• —No percentage saved and no efficiency figure. Nothing has been measured against a control, so any number would be a guess wearing a decimal point.
• —No satisfaction score. No survey has been run.
• —No quote from the mayor or from any official. Nobody has been asked to give one for this page.
• —No customer logos and no city count. There is one city, it is named above, and it is ours.