Defines the abstract base class for all registry code adapters (BaseRegistryAdapter) and weights adapters (BaseWeightsAdapter). Implements a clean plugin architecture that allows adding new model sources (GitHub, HuggingFace, local FS, OpenUBA Hub) without modifying existing code. The adapters handle listing, downloading, and verifying model integrity. This pattern is directly reusable in any project needing a pluggable resource registry.

Implements an LLM-powered chat service that supports multiple providers (Ollama, OpenAI, Claude, Gemini) with real-time SSE streaming. It dynamically injects system context by querying the database for current anomaly counts, model stats, and page-specific data, enabling context-aware responses. The service abstracts provider details behind a unified interface and handles streaming, error recovery, and token limit management. This component can be extracted and reused in any application needing a customizable, context-aware AI assistant.

A universal chart rendering component that supports nine visualization backends (matplotlib, seaborn, plotly, bokeh, altair, plotnine, datashader, networkx, geopandas). It dynamically loads the required UMD library from a CDN and renders the chart in an iframe or SVG/PNG. Includes download utilities and a fallback mechanism. This component is fully self-contained and can be dropped into any React project to provide flexible, user-configurable chart rendering.

Provides the OpenUBAClient class, a comprehensive Python SDK that wraps all platform REST endpoints (model registration, training/inference jobs, dataset management, visualizations, dashboards, feature store, experiments, pipelines, UBA queries). Handles authentication, session management, caching, error handling, and automatic retries. Uses environment variables for configuration and supports both sync and async operations. This client can be extracted and used as a template for building SDKs for other REST APIs.