org.llm4s.media

A MediaType statically known to be audio, so audio APIs can require one.

A MediaType statically known to be an image, so image APIs can require one.

The broad kind of a media type - the part of a MIME type before the slash.

This is the discriminator multimodal code branches on: extraction decides whether to decode a file as pixels, samples or text; a provider decides whether it can accept the payload at all. Modelling it as a type rather than as mimeType.startsWith("image/") keeps that decision exhaustive and testable.

A media type: its MIME string, its canonical file extension, and its MediaCategory.

This is the shared vocabulary the multimodal modules speak. Before it existed, llm4s-core carried three overlapping enumerations of the same handful of image formats - imagegeneration.ImageFormat, imageprocessing.ImageFormat and imageprocessing.MediaType - which could not be passed to one another despite being structurally identical, and RAG's media extraction matched on raw MIME prefixes instead.

llm4s-media holds the vocabulary only: no I/O, no sniffing, no third-party dependencies. Deciding what a file actually is from its bytes needs Tika, which lives in llm4s-rag; that code produces a MIME string and resolves it here via MediaType.fromMimeType.