pip install retriever-coreGitHub →
Start with a camera.
Run the live webcam Flow graph from Retriever Hub. It detects red and blue objects, then streams the result to Rerun on your machine.
pip install "retriever-core[demo]"
retriever demo webcam --seconds 60 --visualize rerun --refreshThe command fetches openretriever/webcam-demo, opens your webcam, and logs detections to Rerun. Runtime only, no demos? pip install retriever-core.
Run the camera demo
Load the live webcam color-detection Flow graph from Hub and view its detections in Rerun.
Learn the concepts
Flows, clocks, and sync policies — and how one graph runs in-process for debugging or on a backend for real.
Debug and replay
Render the graph, step it locally with normal breakpoints, record a run, and replay exactly what each node consumed.
Explore robot examples
Perception, memory, language, and simulation examples in GoldenRetriever, plus reusable payload types through the Hub.
Design principles
Compose a closed-loop hierarchy
Planning, memory, monitoring, skills, and control make decisions at different horizons. Retriever connects those levels in one hierarchical graph, with feedback and timing kept explicit.
Keep the closed loop explicit
Perception, memory, planning, skills, monitoring, and control stay visible as one causal graph instead of disappearing into callback glue.