RigSpark

About RigSpark

RigSpark is a free, open-source tool for people who want to run large language models on their own computers. It detects your RAM, VRAM or unified memory, memory bandwidth and free disk space, then gives each model in its catalog of 254 a verdict of yes, slow or no, along with an estimated tokens-per-second range and the reason behind the call.

The advice works fully offline and is deterministic. Figures that have no sourced data are shown as unknown instead of guessed. The command line offers a ranked recommendation list, a single-model check that can serve as a CI gate, and a command that pulls, verifies and serves the chosen model on localhost. It sits on top of runtimes such as Ollama, llama.cpp, MLX and LM Studio rather than replacing them.

Beyond sizing, the project includes a loopback-only browser workspace, a hardware score, context-window-aware memory math, reusable agents and skills, and MCP tool connectors. It is written in Rust, MIT licensed, and installs through Homebrew, Cargo, a container image or prebuilt archives for macOS, Linux and Windows.