Qlib

Qlib is Microsoft’s AI-oriented quantitative investment platform with 49,500+ stars on GitHub, covering the full quant research pipeline from data and model training to backtesting and production strategies.

Overview

Quant research usually means stitching together data feeds, feature scripts, model code and a backtester that were never designed to fit. Qlib replaces that patchwork with one platform built for machine learning from the start. It supports diverse modeling paradigms — supervised learning, market-dynamics modeling and reinforcement learning — over point-in-time financial data, and ships with production-grade feature sets such as Alpha158 and Alpha360 plus a model zoo spanning LightGBM, LSTM, Transformer and other architectures. The workflow runs end to end: data preparation, model training and evaluation, strategy construction, backtesting with portfolio analysis, and on toward live deployment. Qlib also integrates Microsoft’s RD-Agent to automate parts of the research-and-development loop, letting agents propose and test factors and models. Developed by Microsoft Research and released under the MIT license, it remains actively maintained.

Key Features

• End-to-end quant workflow – data, features, models, strategies and backtests in one framework.
• Multiple ML paradigms – supervised learning, market-dynamics modeling and reinforcement learning.
• Ready-made feature sets – Alpha158 and Alpha360 expressions for stock prediction research.
• Model zoo and benchmarks – LightGBM, LSTM, Transformer, TCN and more, with comparable results.
• Backtesting engine – strategy simulation with portfolio, risk and performance analysis.
• RD-Agent integration – automated factor and model research driven by LLM agents.
• Modular and extensible – swap data handlers, models and strategies without rewriting the pipeline.

💳 Pricing

Qlib is free and open source under the MIT license. You run it on your own infrastructure and bring your own market data; Microsoft charges nothing for the platform itself.

Who Is It For

Quantitative researchers, fintech teams and machine-learning engineers who build data-driven trading strategies and want a rigorous, reproducible platform instead of disconnected notebooks and ad-hoc backtests.