Read the situation.
The simulation gives Jev the current conditions, crowding, incidents, available resources, the actions it can take and a 15-second forecast for each.
AN INDEPENDENT EXPERIMENT / BY THOMAS KANZE
Built in the jungle.
Run by Jev.
A little curiosity, a complicated city, and a model I wanted to put to work.
WHY I BUILT THIS
What happens when you give AI something to run?
I’m Thomas, a curious indie hacker living in the jungle. I like building things to find out what’s possible.
What caught my attention about Jev was its potential for everyday problems: deciding what needs attention, choosing the next action, and helping software make sense of a messy situation. Small decisions, made over and over, can make a real difference.
That’s where Run by Jev came from. I wanted to give those decisions a world you could see and interact with. A subway network felt like the perfect place to start: connected, constantly moving, and full of tradeoffs.
So I built this experiment to showcase Jev’s capabilities. Give it a system, put it under pressure, and watch what happens. The recoveries are interesting. So are the mistakes.
Thomas Kanze@thomaskanze on X ↗Jev is TypeSafe AI’s first System One model. It understands natural language and returns structured decisions and probabilities that software can use directly.
Think of routing a support request, ranking what needs attention, or choosing an action from a set of possibilities. Your software supplies the context and available options. Jev makes a judgment; the software checks the result and handles the next step.
That focus on useful, everyday decisions is what made me want to build with it.
THE EXPERIMENT / NEW YORK & LONDON
A city makes the choices visible.
Make the next move.
Jev chooses whether to dispatch a reserve, reroute riders, add capacity, or maintain service. The code validates and applies that choice.
Follow the consequences.
Watch the network respond. Explore recorded choices and measured outcomes in Data, or save a run in History to replay it later.
WHAT YOU’RE WATCHING
Real city geography.
Simulated networks.
New York uses MTA station and service data; London uses TfL station coordinates and Tube route sequences. Jev chooses responses through the TypeSafe API. Passenger demand, capacity, timing, and disruptions are simulated; Jev does not operate actual trains.
This is an experiment in what the model can do. The Data page lets you inspect its choices, latency, costs, and results alongside a rule-based operator. The results include failures and ties, because those belong in the story too.
Explore the evidence ↗CURIOSITY GOT US HERE