Skip to content
HUIAHUB

AlexsJones/llmfit

llmfit

Hundreds of models & providers. One command to find what runs on your hardware.

Metrics updated Established
View on GitHub (opens in a new tab)

GitHub metrics

Stars
38K
growth after 7 days of data
Forks
2.4K
30-day growth
—
Open issues
68

The chart appears once the repository has been checked on two or more days.

What it does

llmfit is a terminal tool, written in Rust, that inspects your hardware and tells you which open-source large language models it can run well.

It detects CPU cores, system RAM, GPUs, VRAM and unified memory (NVIDIA CUDA, Apple Silicon, AMD ROCm and Intel OneAPI), then estimates memory use and tokens per second for each model across quantizations. Each model gets scores for quality, speed, fit and context, plus a fit verdict that runs from Perfect to Too Tight. The default interface is an interactive TUI. A classic CLI mode, a web dashboard and a REST API (/api/v1/system and /api/v1/models) are also available. llmfit works with Ollama, llama.cpp, MLX, Docker Model Runner and LM Studio, and can download a chosen model through them.

Speed figures come from a memory-bandwidth formula unless a measurement exists. A newer benchmark feature measures real tok/s on your machine and lets you submit the result to the project as a pull request from the TUI, so later releases carry measured numbers for matching hardware. The tool runs on macOS, Linux and Windows and is MIT licensed. The latest release is v1.1.17.

Why it matters

Choosing a local model usually means guessing whether it fits in memory and how fast it will run. llmfit does that arithmetic from the hardware it detects, accounting for quantization, context length, multi-GPU setups and Mixture-of-Experts models, where only active experts count toward per-token memory. Each estimate carries a confidence label, and its inputs can be checked with `llmfit info`, so a reader can see why a model was ranked where it was.

Main features

  • Detects CPU, RAM, GPUs, VRAM and unified memory automatically
  • Estimates memory footprint and tokens per second across quantization formats
  • Scores models on quality, speed, fit and context
  • Interactive TUI by default, with a CLI mode and a web dashboard
  • REST endpoints `/api/v1/system` and `/api/v1/models` through `llmfit serve`
  • Downloads models via Ollama, llama.cpp, MLX, Docker Model Runner or LM Studio
  • Benchmarks real tok/s locally and submits results as a pull request
  • Machine-readable JSON output from `llmfit recommend` for scripts

Who it's for

  • People choosing which local LLM to download for a laptop or workstation
  • Ollama, llama.cpp, MLX and LM Studio users who want a fit check before pulling a model
  • Teams that want hardware-to-model data in scripts through JSON output or a REST API

Strengths

  • Covers NVIDIA, AMD, Intel and Apple Silicon hardware on three operating systems.
  • Handles multi-GPU memory and Mixture-of-Experts models with expert offloading.
  • Labels each estimate (for example measured_local, calibrated or estimated) so the source of a number is visible.
  • Many install routes, from Homebrew and uv to a shell script and a Docker image.

Limitations

  • Speed and fit numbers are formula-based estimates unless someone has measured that hardware.
  • The docs say generic Mixture-of-Experts estimates have missed before, putting gpt-oss-120b about 2.6x low.
  • Community measurements reach users only after a pull request is merged and released.
  • A release may publish an unsigned Windows binary if the signing job fails.

Getting started

Prebuilt binaries ship through Homebrew, an install script, uv and a Docker image. Use one of these:

brew install AlexsJones/llmfit/llmfit
curl -fsSL https://llmfit.axjns.dev/install.sh | sh
uv tool install -U llmfit
docker run -it --rm ghcr.io/alexsjones/llmfit --tui

Requirements: macOS (Apple Silicon or Intel), Linux (x86_64 or ARM64) or Windows (x86_64).

Always follow the official instructions in the README on GitHub. This overview is our summary, not a copy.

Sources