GoldenRetriever Hub Module v1¶
GoldenRetriever is the first applied robotics Hub module for Retriever. The core
runtime package is retriever-core and imports as retriever; Golden's reusable
robot-facing payloads are exposed through the Retriever Hub module manifest in
pyproject.toml.
Core boundary¶
Start with the core visual quickstart if you are new to Retriever: https://openretriever-docs.pages.dev/getting-started/visual-quickstart/. The core runtime provides Flow, Pipeline, registry, IR, and Hub mechanics; Golden provides applied robot-facing payloads and maintained examples.
Fast proof from source checkout¶
Run these from the GoldenRetriever repository:
The Hub-pack smoke command does four things without network access:
- reads the local
[tool.retriever.module]manifest, - loads representative exports through the runtime Hub loader,
- checks that applied types such as
WorldStateare visible through the unified registry, - round-trips a lightweight
Actionpayload through the exported Arrow helpers.
The visualization command validates a small closed-loop graph to IR and writes a
self-contained HTML artifact under out/.
Typical output:
Golden Hub exports: WorldState, BeliefGraph, Skill, Plan, Trajectory, convert_to_arrow, convert_from_arrow
Registry WorldState: _retriever_hub...WorldState
Constructed WorldState: ['cup']
Constructed Plan skills: ['pick']
Arrow round-trip: Action OK
Hub reference: hub.use("openretriever/golden-retriever:WorldState")
Graph proof: run `pixi run demo-pipeline-html-viz` to validate and render an IR HTML artifact.
Hub reference¶
Golden exports load the same way as any other Hub module:
from retriever import hub
WorldState = hub.use("openretriever/golden-retriever:WorldState")
Plan = hub.use("openretriever/golden-retriever:Plan")
convert_to_arrow = hub.use("openretriever/golden-retriever:convert_to_arrow")
Golden does not need a separate runtime package for this path. Users install the runtime once, then load domain packs through Hub.
What this proves¶
This page is intentionally narrower than a full robot demo. It proves the release-critical extension boundary:
- the manifest can be loaded from a standard
src/layout, - Golden exports appear as Hub-loaded objects,
- applied payloads register in the runtime registry,
- conversion helpers are exported with the pack,
- graph validation and visualization remain runtime-owned.
For richer examples, continue with perception, memory, language, composition, and simulation lanes in the Example Guides section.