Skip to content

WorldForge

A harness framework for building world-model-based workflows for physical AI systems.

WorldForge is the application builder's counterpart to model-training stacks like Stable World Model: it helps roboticists and physical-AI builders compose, evaluate, and benchmark workflows built on top of world models — so they can pick the best provider and configuration for a task — rather than train the models. The whole framework is organized around one backbone loop: planning and scoring action candidates with an action-conditioned predictive world model, in latent space.

Get started Read the introduction

Where to go next

  • Quick Start

    Install WorldForge, create a world, and run the mock provider end to end.

    Quick Start

  • CLI Reference

    Look up an exact CLI command or optional-runtime smoke.

    CLI Reference

  • Architecture

    See module responsibilities and the planning pipelines.

    Architecture

  • Provider Authoring

    Add or promote a provider adapter against the capability contracts.

    Provider Authoring Guide

  • Documentation Map

    Follow the complete reader path across every section.

    Documentation Map

  • Evidence And Evaluation

    Understand how claims map to deterministic, reproducible evidence.

    Evaluation