# Retriever > Retriever (PyPI target `retriever-core`, import package `retriever`) is a typed dataflow runtime for closed-loop robot systems. It lets users author typed Flow graphs with explicit clocks and per-edge sync policies, inspect them as graphs, debug them in-process with `pipe.step()`, then run the same graph on multiprocessing or dora backends with recording and replay. ## Agent First Path If you are an AI agent working in the source repo, do this before editing code: ```bash pixi install pixi run demo-webcam-detection-mock pixi run demo-stepper pixi run docs-tutorial-perception-html ``` Use the mock visual smoke first. Do not start with live camera, dora, or multiprocessing unless the user specifically asks for that backend. ## Critical Facts - Python 3.11+. - PyPI distribution target: `retriever-core`. - Import package: `retriever`. - The source checkout is still the canonical path for demos, examples, Rerun visualization, and repository tests. - Define payloads with `@io` from `retriever.flow`. - Override `Flow.step(...)`; do not invent `run(...)` methods for Flow logic. - Edge sync should be explicit with `connect(..., sync=...)` unless a default sync policy is intentionally configured. - Common backends are `in-process`, `multiprocessing`, and `dora`. - Use record/replay when behavior depends on live sensor timing. - GoldenRetriever is the maintained reference examples layer: maintained robot-facing examples, type packs, simulator/visualization lanes, and Hub-pack candidates. ## Canonical Commands ```bash pixi run demo-basic-flow pixi run demo-stepper pixi run demo-perception-stepper pixi run demo-webcam-detection-mock pixi run demo-webcam-detection pixi run docs-tutorial-perception-html pixi run demo-webcam-record pixi run demo-webcam-replay-rrd ``` ## Docs - [Overview](https://openretriever-docs.pages.dev/): recommended path and conceptual map. - [Install](https://openretriever-docs.pages.dev/getting-started/install/): source-checkout setup today; `retriever-core` package target after 0.0.1 resolves. - [Visual Quickstart](https://openretriever-docs.pages.dev/getting-started/visual-quickstart/): deterministic mock color-detection smoke, then live webcam/Rerun. - [Examples and Results](https://openretriever-docs.pages.dev/tutorials/examples-and-results/): commands paired with expected output. - [Debug and Visualize](https://openretriever-docs.pages.dev/tutorials/debug-and-visualize/): graph render, local stepping, Rerun, record, replay. - [Flow](https://openretriever-docs.pages.dev/concepts/flow/): typed Flow authoring. - [Time and Sync](https://openretriever-docs.pages.dev/concepts/time-and-sync/): clocks, event streams, and sync policies. - [Runtime](https://openretriever-docs.pages.dev/concepts/runtime/): validate -> IR -> backends; stepping; recording and replay. - [Standard Types](https://openretriever-docs.pages.dev/concepts/standard-types/): runtime-owned type primitives and what belongs outside core. - [Concepts and Lineage](https://openretriever-docs.pages.dev/concepts/lineage/): relation to FRP and synchronous dataflow. - [Golden Hub Proof](https://openretriever-docs.pages.dev/ecosystem/golden-packs/): GoldenRetriever as a manifest-declared type pack loaded through Retriever Hub. - [Modules](https://openretriever-docs.pages.dev/ecosystem/modules/): Retriever Hub module loading with `hub.use("org/name:Export")`. - [Composable Pipelines](https://openretriever-docs.pages.dev/ecosystem/composable-pipelines/): reusable pipelines as inspectable graphs or Flow stages. - [Publishing](https://openretriever-docs.pages.dev/ecosystem/publishing/): make a repo hub-loadable with `[tool.retriever.module]`. ## Companion Sites - [Core Source](https://github.com/openretriever/retriever): runtime implementation and source-checkout demos. - [Golden Source](https://github.com/openretriever/golden-retriever): robot-facing examples, type packs, simulator lanes, and public-surface checks. - [Landing page](https://openretriever.org/): compact public front door. - [GoldenRetriever](https://retriever-space.pages.dev/): robot-facing reference examples, type packs, simulator/visualization lanes, and Hub-pack candidates.