AMD's Lemonade SDK 10.3: 101MB to 7MB by Ditching Electron

AMD's Lemonade SDK just went from 101MB to 7MB. Version 10.3 dumps Electron for Tauri, and that single swap cut binary sizes by roughly 10x. According to Michael Larabel at Phoronix, the pre-built macOS and Windows binaries are now a fraction of what they were. Tauri's a Rust-based framework for building desktop apps with web tech. Read the full announcement for details on how this works. It pulled this off without losing features.

The update also brings OmniRouter, which bundles backend engines for what AMD calls a "true omni-modal" LLM experience. That means simpler interactions across text, image, and audio. Users can now pick specific Llama.cpp versions or flip on auto-updating. There's also support for switching between ROCm builds: 7.2 stable, 7.12 preview (the new default), and TheRock nightly builds.

Compare that to NVIDIA's toolkit, which needs separate SDKs for different modalities. Lemonade runs entirely on local hardware, exposes OpenAI-compatible API endpoints, and works across AMD CPUs, GPUs, and NPUs. No CUDA lock-in. For 7MB, you're getting a lot. AMD's GAIA SDK takes a similar approach for agent development.