retrain-pipelines is an Apache 2 pip-installable, all-in-one python framework that embeds its own DAG engine and WebConsole, eliminating external orchestration dependencies. Its core innovation is a nomadic execution model: the same pipeline can run seamlessly across local machines, distributed environments, or directly inside notebooks (Jupyter/Colab), with both execution backend and UI co-located in a single runtime if needed.

Pipelines are defined as minimal decorated Python functions and executed as parameterized runs, separating execution parameters (data, hyperparameters) from worker parameters (compute, parallelism, distribution). This enables reproducible yet flexible retraining scenarios without redefining the pipeline.

The framework natively supports runtime-expanded DAGs (e.g., distributed subgraphs whose cardinality is decided at execution time), while maintaining full observability via its built-in WebConsole (live DAG, traces, timelines).

Unlike generic orchestrators, retrain-pipelines tightly integrates distributed retraining logic, execution control, and observability in a single portable unit, optimized for continuous learning workflows rather than static job scheduling.