# Counterworld

> Counterworld creates stateful software worlds with simulated users, controllable time, forks, replay, and verifiable outcomes for training and evaluating AI agents.

Counterworld is simulation infrastructure for AI agents. A world contains executable APIs, business rules, persistent state, history, user actors, and delayed consequences. Agents interact through native interfaces while researchers control time, fork state, replay runs, and evaluate observable outcomes.

Use `https://counterworld.dev/` as the canonical domain. Prototype pages and internal visual experiments are intentionally excluded from the resources below.

## Product

- [Counterworld overview](https://counterworld.dev/): What Counterworld is and why agents need living software worlds.
- [Platform](https://counterworld.dev/platform.html): How worlds, controllable time, forks, replay, and outcome evaluation work together.
- [Worlds](https://counterworld.dev/worlds.html): Maintained SaaS worlds, internal-system replicas, and custom world definitions.
- [Solutions](https://counterworld.dev/solutions.html): Agent training, specialization, long-horizon evaluation, regression testing, and production replicas.

## Documentation

- [Documentation guide](https://counterworld.dev/docs/llms.txt): Curated technical documentation for agents and developers.
- [Documentation home](https://counterworld.dev/docs/): Counterworld concepts and interface map.
- [Quickstart](https://counterworld.dev/docs/quickstart.html): Create a world, connect an agent, advance time, fork state, and evaluate outcomes.
- [Interface contracts](https://counterworld.dev/docs/reference.html): Agent, control, inspection, simulation, and evaluation surfaces.

## Access

- [Request early access](https://counterworld.dev/early-access.html): Apply to train or evaluate an agent in Counterworld.