PyCaret
PyCaret is a free, open-source, low-code Python tool for automating machine-learning workflows.
What PyCaret is
PyCaret is a free, open-source, low-code Python tool for automating machine-learning workflows. Its experiment APIs cover classification, regression, clustering, anomaly detection, and time-series forecasting. The engine can fit experiments, compare and tune models, make predictions, and save or load models. Fitted results can be carried forward as portable scikit-learn pipelines, with forecasting pipelines for time-series work. The core engine installs from PyPI with `pip install pycaret`, and Python 3.11, 3.12, and 3.13 are listed as supported. The default path runs on CPU; GPU-accelerated estimators are available when the required libraries are installed. Forecasting, explainability, and dashboard dependencies are optional extras. An optional React dashboard and FastAPI backend can be self-hosted for multi-user and deployment workflows. Saved pipelines can be served behind FastAPI, and an optional PyCaret server offers an HTTP prediction endpoint. The backend can be connected to existing authentication and is JWT-friendly. PyCaret is MIT-licensed, with no paid or enterprise plan stated. Its FAQ directs questions to GitHub Discussions and bug reports to GitHub Issues.
Good
- Covers five machine-learning task types.
- Fitted outputs can be portable pipelines.
- Supports Python 3.11, 3.12, and 3.13.
- Free, open-source, and MIT-licensed.
Watch out
- GPU estimators require relevant libraries.
- Dashboard and forecasting dependencies are optional extras.
- Existing 3.x code may need to stay on 3.x.
Who it suits
PyCaret suits Python users who want low-code tools for machine-learning experiments across several task types. Its optional self-hosted dashboard and server may suit teams building multi-user or deployment workflows.
Verdict
PyCaret brings experiment setup, model comparison, and pipeline handling into one Python tool. Note the 3.x API compatibility guidance before moving existing code to 4.0.
Plans and prices
From PyCaret's own pricing page, checked 7 Oct 2026.
| Plan | Price | What it gives you |
|---|---|---|
| PyCaret | Free | MIT-licensed open source; no paid tier or enterprise plan stated |
| PyCaret open-source | Free | No paid tier · MIT open-source engine · self-hosted |
The facts
Each one read from PyCaret's own pages, with the date it was checked.
What it does
- Anomaly detection
- AnomalyExperiment uses pyod adapters including Isolation Forest, KNN, COPOD, ECOD, LOF, PCA, MCD, ABOD, CBLOF, HBOS, OCSVM, SOS, and SOD.checked 15 Oct 2026
- Backend API
- The optional pycaret-server backend is a FastAPI service with CRUD, run orchestration, WebSockets, pipeline deployment, and plot endpoints.checked 15 Oct 2026
- Classification
- ClassificationExperiment handles categorical targets and includes metrics such as accuracy, AUC, recall, precision, F1, kappa, and MCC.checked 15 Oct 2026
- Clustering
- ClusteringExperiment offers KMeans, affinity propagation, mean shift, spectral, hierarchical, DBSCAN, OPTICS, Birch, KMode, and HDBSCAN adapters.checked 15 Oct 2026
- Compatibility break
- Version 4.0 removes the 3.x module-level functional API and requires the new Experiment object-oriented API; mixing 3.x and 4.0 is unsupported.checked 15 Oct 2026
- Compatibility limit
- PyCaret 4.0 removes the 3.x module-level functional API, so the FAQ advises pinning 3.x for existing code and using 4.0 for new work.checked 7 Oct 2026
- Current maturity
- The GitHub repository labels the 4.0 line as work in progress and says the 3.x line is frozen at 3.4.0.checked 15 Oct 2026
- Dashboard
- An optional React dashboard and FastAPI backend can be self-hosted for multi-user and deployment workflows.checked 7 Oct 2026
- Deployment
- Models can be saved as portable scikit-learn pipelines, served behind FastAPI, or deployed through the optional pycaret-server layer.checked 15 Oct 2026
- Documentation depth
- The official site includes concepts, tutorials, preprocessing references, function guides, FAQs, a cheat sheet, and generated API reference.checked 15 Oct 2026
- Explainability and plots
- The plotting modules expose diagnostic charts for classification, regression, features, clustering, anomalies, time series, and exploratory analysis.checked 15 Oct 2026
- Forecasting
- TimeSeriesExperiment is built on sktime and supports models such as ARIMA, AutoARIMA, exponential smoothing, ETS, Theta, STLF, Croston, and reduced-regressor adapters.checked 15 Oct 2026
- Hardware
- The default path runs on CPU, and GPU-accelerated estimators are available when the relevant libraries are installed.checked 7 Oct 2026
- Hyperparameter tuning
- tune_model performs configurable iterative hyperparameter searches and returns tuned pipelines and parameters.checked 15 Oct 2026
- Installation
- The documented installation is pip install pycaret, with dashboard, explainability, and forecasting extras available.checked 15 Oct 2026
- Integrations
- PyCaret wraps or interoperates with scikit-learn, XGBoost, Microsoft LightGBM, spaCy, sktime, pyod, FastAPI, React, and Plotly.checked 15 Oct 2026
- Known limitations
- The 4.0 changelog lists incomplete plot rewrites, legacy internals still being replaced, and no Python 3.14 support.checked 15 Oct 2026
- Licence
- PyCaret 4.0 licenses the engine under MIT and the control-plane components under BUSL-1.1 with a later Apache-2.0 change date and an additional-use grant.checked 15 Oct 2026
- Limit: comparative performance
- A 2023 healthcare study reported PyCaret accuracy below AutoGluon on several Cleveland, combined, and Hungarian dataset tests, with Hungarian unreduced accuracy around 68%.checked 15 Oct 2026
The company
- Company description
- LinkedIn describes PyCaret as an open-source, low-code machine-learning library and a nonprofit software-development organization.checked 15 Oct 2026
- Company location
- LinkedIn lists Toronto, Canada as PyCaret's headquarters and primary location.checked 15 Oct 2026
- Company size
- LinkedIn lists PyCaret's company size as 11–50 employees.checked 15 Oct 2026
- Founded
- LinkedIn lists PyCaret as founded in 2020.checked 15 Oct 2026
- Headquarters
- Toronto, Canadachecked 28 Sep 2026
Alternatives to PyCaret
- 1
BigML
Free planWebAPImacOSSelf-hosted7.7From $10,000 a year - 2
Oracle WebCenter Content
AndroidiOSWebAPI7.7$3,450 one-time - 3
Weka
Free planOpen sourceAPILinuxmacOSSelf-hosted7.5Free plan - 4
KNIME Analytics Platform
Free planLinuxmacOSWindows7.1From $19 a month - 5
MLJAR AutoML
Free planWebLinuxmacOSSelf-hosted7.1From $20 a month - 6
Obviously AI
Free planWebAPI7.1Free plan