kalosm

Why we included this project

Kalosm is a Rust crate ecosystem that puts pre-trained language, audio, and image models behind one async interface, so a single dependency can drive a Llama chat session, Whisper transcription, or semantic search instead of wiring several SDKs together. It runs models on the user's own machine by default, which keeps inference private and workable offline. The strongest part is structured generation: you derive typed parsers from your Rust types, and the model is forced to emit JSON, HTML, or XML that matches your schema instead of free-form text. Fusor, the bundled WGPU runtime, handles quantized inference on native GPUs or in the browser. If you already write Rust, that means model features arrive as ordinary library calls rather than a separate server to manage.