Select open-source AI models that fit your hardware automatically
Choosing an open-source language model that actually runs on your hardware from hundreds of options remains difficult. This tool automatically scans your CPU, RAM, and GPU memory to score executable models. A single command in the terminal produces a ranked list of models that run smoothly on your equipment.
Benchmarks that replace estimates with actual measurements
A recent update added the ability to run models directly on your computer to measure tokens per second and share those results with the project. You can run benchmarks directly from the terminal interface and submit the results, allowing other users with the same hardware to see actual measured figures rather than estimates. This process requires no external account or separate CLI tool; data is stored locally first and reflected in the next release.
Scoring the compatibility between hardware and models
The tool automatically detects the system's CPU core count, system RAM, GPU type, VRAM capacity, and unified memory architecture. Based on these detected specifications, it analyzes the model's parameter count, context length, and quantization format (GGUF, AWQ, GPTQ, EXL2) to calculate memory usage and expected speed. Each model is scored and ranked across four dimensions: suitability, speed, quality, and context handling capability. Speed estimates are based on memory bandwidth models and actual community measurement data, and you can check which assumptions are reflected in the numbers using the llmfit info command.
Usage across terminal, web, and API environments
By default, it provides a lightweight terminal interface (TUI) with no dependencies, but it can also run as a web dashboard or REST API server. In the TUI, you can navigate and filter the model list using keyboard shortcuts and view the community benchmark leaderboard. If you need automation pipelines or dashboard integration, you can retrieve JSON-formatted data via the /api/v1/system and /api/v1/models endpoints. You can easily deploy it in container environments as an interactive mode or headless API server using the Docker image.
Things to check before use
This tool recommends models by combining hardware-based estimates with community-contributed data, but it does not guarantee that every model will work perfectly in your environment. While supporting MoE (Mixture-of-Experts) architecture is a strength, you must verify compatibility with specific runtime providers (Ollama, llama.cpp, MLX, etc.) yourself. Windows binaries are signed only if the code signing process completes successfully, so unsigned artifacts may be distributed; in security-critical environments, you must verify executable signatures. Additionally, this program does not send information to external networks unless explicitly requested by the user (such as model downloads or leaderboard lookups).
