Right-size LLM models to your hardware

Terminal tool that detects your system specs, scores hundreds of models across quality, speed, fit, and context, and tells you which ones will actually run well on your machine.

Get Started Command Reference

\$ llmfit

System: 62GB RAM | 14 cores | NVIDIA RTX 4090 (24GB VRAM)

Qwen/Qwen2.5-Coder-7B Score: 94.5

Mistral-7B-Instruct Score: 92.1

Llama-3.1-8B-Instruct Score: 91.8

Quick Start

Get llmfit running on your system in under a minute

Install using your preferred package manager: ```bash macOS theme={null} brew install llmfit ``` ```bash Windows theme={null} scoop install llmfit ``` ```bash Linux theme={null} curl -fsSL https://llmfit.axjns.dev/install.sh | sh ``` ```bash From source theme={null} git clone https://github.com/AlexsJones/llmfit.git cd llmfit cargo build --release ``` Launch llmfit to see which models fit your hardware: ```bash theme={null} llmfit ``` The TUI displays your system specs at the top and ranks all models by fit score. Use arrow keys or `j`/`k` to navigate, `/` to search, and `q` to quit. Your system specs (CPU, RAM, GPU name, VRAM, backend) appear at the top. Models are listed in a scrollable table sorted by composite score, with columns for: * **Score**: Composite score (0-100) across quality, speed, fit, and context * **TPS**: Estimated tokens per second * **Quant**: Best quantization for your hardware (Q8\_0 to Q2\_K) * **Mode**: Run mode (GPU, CPU+GPU, CPU, MoE) * **Mem**: Memory usage percentage * **Context**: Maximum context length Use keyboard shortcuts to filter results: * Press `f` to cycle fit filters (All, Runnable, Perfect, Good, Marginal) * Press `/` to search by name, provider, or use case * Press `d` on a selected model to download it via Ollama or llama.cpp ```bash theme={null} # Or use CLI mode for scripting llmfit fit --perfect -n 5 llmfit recommend --json --use-case coding --limit 3 ``` Models marked with a green ✓ in the **Inst** column are already installed via Ollama, llama.cpp, or MLX.

Explore by Topic

Deep dive into llmfit's features and capabilities

Understand hardware detection, scoring algorithms, and fit analysis Master the interactive terminal UI with keybindings and themes Use llmfit in scripts with subcommands and JSON output Run llmfit as a node-level API for cluster scheduling Connect with Ollama, llama.cpp, and MLX for model downloads GPU detection for NVIDIA, AMD, Intel Arc, and Apple Silicon

Key Features

Multi-GPU Detection

Detects NVIDIA, AMD ROCm, Intel Arc, Apple Silicon, and Ascend NPUs with aggregated VRAM reporting

Dynamic Quantization

Automatically selects the best quantization (Q8\_0 to Q2\_K) that fits your available memory

Multi-Dimensional Scoring

Ranks models across quality, speed, memory fit, and context window with use-case-specific weights

MoE Architecture Support

Optimizes Mixture-of-Experts models like Mixtral and DeepSeek with expert offloading