Machine learning intent classification powered by ML.NET. Models are trained per environment at deploy time and persisted via the engine-shared workspace data store.

OpenAI core: chat completions, the Responses API, embeddings, moderation, model list, conversations, containers, streaming variants, webhook receiver. The satellite modules (Assistants, Files & Vector Stores, Images, Audio, Batch & Fine-tuning, Realtime) are installed and configured separately.

OpenAI Assistants: threads (incl. streaming), assistant management, run cancellation, and the function-tool callback loop. Configured separately: set this module's own OpenAI API key (and optional organization) in its environment configuration.

OpenAI audio: text-to-speech, transcription (incl. diarized) and audio translation. Configured separately: set this module's own OpenAI API key (and optional organization) in its environment configuration.