The page is set over an illustrative simulated city, rendered in your browser: authored residents live through three compressed days until one of them forms hostile intent, and a perception layer brackets every figure and names the ones that carry the story. It is a synthetic visualization, not footage of real people and not output of the ouro product.
You can’t film rare behavior on demand. So we built a city where you can.
ouro is a sandbox city full of NPCs living their lives. You type what should happen — ouro stages it, records it and labels it, ready to use.
Object detection is solved.
Predicting human intent is next.
A model can’t learn behavior it has never seen, and cameras almost never see the ones that matter: concealment, a handoff, infiltrating a base. Most synthetic pipelines script one actor at a time from a closed list of moves. Nobody interacts.
Two years ago this took a team of operators. Now it takes a sentence.
- 01 · authorYou write the sentence: “Two residents perform a handoff in an alley at 21:00, observed by a patrolling drone.” An LLM builds the scenario.
- 02 · runIt plays out in a city already living its own day.
- 03 · trainLabeled video out, annotated per frame. Change one word, run it again.
No generated footage. We simulate a world and record what happened in it.
Your world. Your stack. Your rules.
Designed from day one to run on-premise and air-gapped, beside the scenarios, simulation tools and environments you already use. Deployed that way, your data stays on your infrastructure.
Not just spotting the incident. Seeing it coming.
No one crosses the fence in this city. The flag is earned by behavior alone.
We co-develop with early design partners. Co-build the benchmark.
Four founders. We built the sim ourselves.Orel Eliad CEO · Ronel Yemin COO · Amit Levy CTO · Aviv Amar CPO