Guild AI is an open-source toolkit that automates and optimises machine learning experiments, with 909 stars on GitHub and a mature, stable codebase.
Guild AI brings experiment discipline to machine learning work without forcing you to rewrite your training code. You run unmodified training scripts, and Guild captures each run’s result as a unique, recorded experiment โ code version, flags, outputs and metrics included. On top of that record-keeping it automates trials with grid search, random search and Bayesian optimisation, so hyperparameter sweeps become one command instead of a folder of ad-hoc logs. Runs can be compared and analysed to understand what actually improves a model, related operations like data preparation and testing can be backed up with the experiment, and results can be archived to S3 or other remote storage for safekeeping and sharing. Guild’s design philosophy is restraint: it stays out of your training code entirely and observes runs from the outside, which makes adoption a matter of minutes rather than a refactor of your project.
Guild AI is free and open source under the Apache-2.0 license, available on PyPI. There is no hosted tier to buy; your only costs are the machines you train on.
Guild AI is for ML engineers and researchers who run many training experiments and need reliable records of what was tried and what worked. It appeals to people who like lightweight tooling: if heavyweight experiment platforms feel like overkill, Guild’s script-first approach keeps your workflow intact while adding tracking and automation.