Skip to content

Retriever Hub

Retriever has three public surfaces that should stay separate but connected:

  • Core runtime: Flow, Clock, Sync Policy, Pipeline, stepping, replay, IR, standard types, and backend execution.
  • Retriever Hub: the reusable pack/module protocol and index for Flows, type packs, transforms, and composed pipelines.
  • GoldenRetriever: the maintained reference examples layer: maintained robot-facing examples, type packs, simulator/visualization lanes, and current manifest-declared Hub pack surface.

Use Hub when a robot system needs to reuse a stable boundary rather than copy code: a Flow class, a pipeline builder, a typed payload pack, or a representation transform.

from retriever import hub
from retriever.flow import Pipeline, Rate, Latest

LidarSlam = hub.use("your-org/lidar-slam:LidarSlamFlow")
slam = LidarSlam(resolution=0.05) @ Rate(hz=10)

hub.use("org/name:Export") returns the exported object itself. That object can be a Flow class, pipeline builder, type, transform, or small utility. Hub is therefore the same distribution path for reusable robot modules, reference example packs, and robot-facing type packs.

  1. Run the core visual quickstart first.
  2. Read Hub packs and modules to understand refs and exports.
  3. Run Golden Packs to prove that robot-facing types load through Hub.
  4. Continue to Golden examples when you want maintained perception, memory, language, simulator, visualization, or robot type-pack lanes.

The important boundary is that GoldenRetriever is not a second runtime package. The runtime is retriever-core and imports as retriever; Golden is the maintained examples site and pack-candidate surface built on top of it.