Skip to main content
The catalog is two artifacts that stay in sync:
  • src/worldflux/curated.py — the Python source the CLI reads at worldflux init --recipe <id> and worldflux curated * time.
  • web/lib/catalog-data.ts — the TypeScript copy the dashboard’s Catalog tab renders.
Both files describe the same adapter set. Keep them aligned when changing the catalog; the Python source is the CLI source of truth and the TS copy is what the dashboard renders.

​
Entry shape

type CatalogAdapter = {
  id: string;
  name: string;
  description: string;
  provider: string;
  providerSlug: string;
  runtime: string;                 // e.g. "local" | "replicate" | "modal" | "nvidia-cloud"
  tier: "active" | "historical" | "demo";
  modelFamily: string;
  endpointTypes: string[];         // "rollout" | "train" | "eval" | "embed" | "predict" | "serve" | "generate"
  minimumVramGb: number | null;
  minimumCuda: string | null;
  upstreamLicense: string;
  version: string;
  smokePrompt: string;
  metricKeys: string[];
  outputArtifacts: string[];
  evalBenchId: string | null;
  knownIssues: string[];
  supportLevel: "production-backed" | "production-disabled" | "experimental-not-shipped";
  supportNote: string;
};
Field notes:
FieldPurpose
idWhat you pass to --recipe. Matches the directory under src/worldflux/templates/ for adapters that ship a template.
runtimeThe default runtime plugin to use. Today the catalog points at local or replicate; deployable adapters can also expose modal as a BYOK deploy target.
tieractive if CI runs against it, historical if it loads but no longer gets fixes. The dashboard pill is green / red.
endpointTypesWhat the adapter can do. Used by the dashboard’s filter + the recipe’s task picker.
minimumVramGbObserved floor on a clean install. Lower works for tiny modes; do not assume.
minimumCudaRequired CUDA toolkit version, if any. null for CPU-friendly adapters.
evalBenchIdThe eval bridge wired in for this adapter (libero, robocasa, vjepa_embed, minecraft_offline).
metricKeysThe metric names the adapter writes into manifest.metrics. Used by worldflux compare to know what to diff.
outputArtifactsFilenames the adapter promises to commit. The manifest writer rejects runs that miss them.
deploy_targets (Python source)Explicit remote serving surfaces available from worldflux curated deploy; entries are not production-backed runtime support unless adapter metadata says so.
smokePromptThe prompt or seed used by the smoke test.
knownIssuesShort notes the dashboard surfaces in a yellow callout.
supportLevel (dashboard copy)Public support claim shown to users. Catalog visibility is not the same as production-backed execution.

​
What ships today

idTierRuntimeBenchPublic support
cosmos_predict25activereplicaterobocasaproduction-disabled
dreamer4activelocalminecraft_offlineexperimental-not-shipped
internvla_m1activelocalliberoexperimental-not-shipped
openpiactivelocalliberoproduction-disabled
smolvla_450mactivelocallerobotexperimental-not-shipped
gr00t_n1_7activelocalliberoproduction-disabled
pi07activelocalliberoexperimental-not-shipped
vjepa2activelocalvjepa_embedexperimental-not-shipped
openvla-libero-pro-lightingdemomodallibero-proproduction-disabled
openvla-libero-pro-object_posedemomodallibero-proproduction-disabled
gr00t-n17-isaac-lab-25taskdemonvidia-cloudgr00t-isaac-labproduction-disabled
cosmos-predict-25-isaac-sim-drift-5framedemomodalcosmos-predict-driftproduction-disabled
diamondhistoricallocal(not wired)experimental-not-shipped
dreamerv3historicallocal(not wired)experimental-not-shipped
stormhistoricallocal(not wired)experimental-not-shipped
tdmpc2historicallocal(not wired)experimental-not-shipped
vjepa2_vitlhistoricallocalvjepa_embedexperimental-not-shipped

​
Templates and deploy targets

Serving templates live under src/worldflux/templates/ for the adapters that support BYOK deploy or dry-run demo flows:
  • cosmos_predict25/
  • openpi/
  • openvla_libero_pro/
  • vjepa2_vitl/
These templates power BYOK deploy/run-remote flows: 
worldflux curated deploy cosmos_predict25 --runtime modal
worldflux curated run-remote cosmos_predict25 generate
worldflux curated status cosmos_predict25
worldflux curated undeploy cosmos_predict25
For local installed adapters, use worldflux use <id> followed by the env tree: 
worldflux use openpi
worldflux env build openpi
worldflux curated run openpi rollout

​
Browsing

worldflux curated list
worldflux curated inspect cosmos_predict25
inspect prints the full record, including known issues. The dashboard’s Catalog tab renders the same data with search and runtime/tier filters.

​
Adding an adapter

1

Add to curated.py

Append a CuratedAdapter(...) entry to src/worldflux/curated_registry.py through the src/worldflux/curated.py compatibility facade. Fill every field; the dashboard treats null as “unknown” and renders accordingly.
2

Update the dashboard copy

Mirror the user-facing fields in web/lib/catalog-data.ts so the Catalog tab stays in sync with the CLI source.
3

Run it once

The cutoff for tier: active is one passing CI run plus a documented benchmark id. Until then, set tier: "historical" and evalBenchId: null.
4

Drop a template (optional)

If the adapter has a non-trivial run.py, add it under src/worldflux/templates/<id>/. The init command picks it up automatically.