What it does

ModelFox is a Rust CLI and library stack for training, running, and monitoring ML models without touching a notebook. Point it at a CSV, it featurizes the data, trains linear and gradient-boosted tree models, and writes a .modelfox file. You then load that file in Elixir, Go, JavaScript, PHP, Python, Ruby, or Rust to make predictions. A local web app (modelfox app) serves up model inspection, tuning, and production monitoring dashboards.

The interesting bit

The monitoring layer is baked in, not bolted on. Each prediction gets an identifier; later you log the true value, and the app tracks drift, accuracy, precision, and recall over time. That closed loop—train, predict, log, compare—is usually the hard part of MLOps, and here it is just two API calls and a browser tab.

Key highlights

• CLI training from CSV with automatic feature engineering and model selection
• Prediction libraries for seven languages, all loading the same .modelfox artifact
• Built-in web UI for model inspection, hyperparameter tuning, and production monitoring
• Drift detection and metric alerts via opt-in prediction logging
• Entire stack is a Cargo workspace; builds with nightly Rust

Caveats

• The crates/app directory (the monitoring web app) is source-available and free for testing, but requires a paid license for production use; the rest is MIT
• README mentions v0.8.0 as the latest release, which may or may not still be current
• Requires nightly Rust toolchain to build from source

Verdict

Worth a look if you want low-ceremony AutoML with polyglot deployment and don’t mind the licensing split on the monitoring app. Skip it if you need deep custom architectures or fully open-source production observability.