> ## Documentation Index
> Fetch the complete documentation index at: https://docs.worldflux.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Introduction

> WorldFlux is a BYO-compute control plane for world-model runs, curated adapters, and audit-grade evidence packages.

WorldFlux is proprietary software. Upstream adapter, model, dependency, and
museum-source license metadata in these docs describes third-party materials
only and does not license WorldFlux itself.

WorldFlux records the runs and evidence packages you already produce through local execution or explicitly imported external evidence. It keeps manifests, metrics, logs, artifacts, comparisons, and share links somewhere your team can search and review. The CLI does the work. The dashboard reads the result.

<CardGroup cols={2}>
  <Card title="Quickstart" icon="play" href="/quickstart">
    Install the CLI, sync a sample run, open it in the dashboard.
  </Card>

  <Card title="Concepts" icon="book" href="/concepts/byo-model">
    BYO model, curated catalog, runtime plugins, eval bridges, eval packets, audit evidence.
  </Card>

  <Card title="CLI" icon="terminal" href="/cli/reference">
    Every flag the CLI accepts, with the JSON you can pipe into CI.
  </Card>

  <Card title="Cloud API" icon="code" href="/cloud/api-reference">
    Workspace, project, run, artifact, claim, museum, share, and billing endpoints.
  </Card>
</CardGroup>

## What it is

A read-then-act surface around the `worldflux` CLI. The CLI mutates files on your disk, packages local or remote run outputs, and posts manifests to the cloud when you ask it to. The dashboard reflects those manifests, lets you compare runs, inspect claim/evidence artifacts, mint public share links, and stops there. No hidden compute, no remote training jobs, no surprise side effects.

## What it is not

* A managed GPU service.
* A model marketplace.
* A replacement for the framework you train with.
* A wrapper that hides the CLI behind a button.
* A credentials proxy for private model weights or vendor accounts.

## Where it fits

You already have:

* A training script, a notebook, or a lab pipeline that emits checkpoints, logs, and metrics.
* A WandB / TensorBoard habit that breaks the moment you walk away from the machine.
* A teammate who asks for a link.

WorldFlux turns the run or evaluation output into a manifest/evidence package, uploads it when requested, and gives you a URL. That is the loop.

## Three personas

<AccordionGroup>
  <Accordion title="WM researcher" icon="flask">
    Runs benchmark sweeps on owned compute. Wants run history, artifact diffs, anti-cherrypick eval packets, audit evidence, and citable share links. Reads the manifest, not the chart.
  </Accordion>

  <Accordion title="ML engineer" icon="terminal">
    Productionizes evaluations for a small team. Cares about deterministic runs, quotas, audit trails, members, billing. Reads tables more than charts.
  </Accordion>

  <Accordion title="Solo lab" icon="user">
    Compares curated WM adapters before committing budget. Browses Cosmos-Predict, Dreamer 4, openpi, V-JEPA 2, and historical baselines, opens one or two runs, decides.
  </Accordion>
</AccordionGroup>

## Design register

The interface is a quiet container around CLI activity that already happened. Two registers share the same tokens:

* **Brand surfaces** (landing, login, signup, public share): atmosphere, serif display headlines, one violet accent for product identity.
* **Product surfaces** (dashboard, settings, run detail): dense, table-first, no decoration. Status colors only inside data rows.

See [Design system](/design/system) for the full token sheet.

## Status

WorldFlux is pre-1.0. Public APIs may change between minor versions. The CLI and docs treat `/openapi.json` as the source of truth for cloud clients.