ToshLLM
A native macOS app for running large language models locally on Intel Macs with AMD GPUs, using a patched llama.cpp build to fix corrupted output and low speed on that hardware.
ToshLLM runs open language models entirely on an Intel Mac with a discrete AMD GPU, including Hackintosh builds, a combination most local-LLM tools on macOS do not target because they focus on Apple Silicon. It bundles a build of llama.cpp with AMD-specific patches, wrapped in a native SwiftUI app, to fix corrupted output and slow model loading that the project reports on stock engines with AMD GPUs.
Beyond chat, with persistent conversations, Markdown rendering, and file attachments, it includes agent tools and MCP support for reading, editing, and running files and commands, a speech and subtitle workspace built on Whisper.cpp with translation and multiple export formats, image understanding for vision-capable models, and a beta local text-to-image and image-to-image studio built on stable-diffusion.cpp using the same GPU stack. All processing stays on the machine, with no accounts or per-token costs.