World-Sim · an a4sx module

A company that actually runs. Point your agent at it.

A simulated company your agent can work in. Use it to train an agent, to build software against data that behaves like a real company's, or to try a process change before you make it for real. Orders arrive, simulated staff key them into siloed systems with real lag and mistakes, and one simulated day passes every 30 seconds. Your agent explores it, proposes a change, and gets a measured, signed KPI delta back.

Why you need it: Why you need a running company

Six jobs that need the same thing: a company that behaves like a real one and that you are allowed to break. World-Sim does no reasoning of its own. Your agent brings the ideas; World-Sim runs the company and measures.

How it works: From your agent to a scored idea, in five calls

Every instance is one company on a block of 11 consecutive ports. You reach it with a short-lived token that a4sx issues, get system accounts from the in-world helpdesk, and score candidate automations at the oracle.

Use cases: What you can do with it today

Each card answers one of the jobs above, shows the mechanism behind it and says whether you can do it today.

Pick a world: Nine kinds of organisation

Each template is chosen so the structure your agent must discover is genuinely different. All of them run on the same world engine (a capacitated transformation-graph simulator with mass conservation) and the same 11-port layout, so what you build against one carries to the others.

Worked example · rheinpump: Automate the order clerk, then let the oracle judge it

At Rheinpump, customer orders arrive by email. A simulated clerk re-types them into the ERP days later, sometimes with typos, sometimes not at all. Here is what an agent can do about that.

FAQ: Straight answers, including what it isn't