llmfit solves one of the most frustrating steps in local LLM workflows: figuring out which model will actually run on your hardware before wasting time downloading it. The interactive TUI presents a scored list of compatible models ranked by how well they fit your available VRAM, RAM, and compute capabilities. Each model shows detailed metrics including expected tokens per second, memory requirements, and maximum context length at different quantization levels.
Built in Rust for speed and reliability, llmfit detects your GPU type, VRAM capacity, system RAM, and CPU capabilities automatically. It then cross-references this hardware profile against its database of over 200 models across providers including Ollama, llama.cpp, MLX, Docker Model Runner, LM Studio, and more. The tool has been adopted as a core dependency by Hugging Face's hf-agents extension, which uses llmfit to automatically select the best model for coding agent workflows.
With over 20,000 GitHub stars and MIT license, llmfit has become the standard hardware-model matching tool in the local LLM ecosystem. It pairs naturally with llmserve, a sister project that handles serving the selected model. The tool fills a gap that no other listed tool addresses: the pre-download decision of whether a model will actually perform well on your specific hardware configuration.