Multi Armed Bandit Python Packages

baytune

A simple, extensible library for developing AutoML systems

pybandits

Python library for Multi-Armed Bandits

orts

Odds-Ratio Thompson Sampling: bandits and A/B tests whose baseline keeps moving. Paper: arXiv 2609.19709

thompson

Thompson is a Python package for multi-armed bandit problems and adaptive decision-making. It supports Thompson Sampling, UCB, and random selection, with applications in A/B testing, recommendations, advertising, experimentation, and resource allocation.

mabalgs

:bust_in_silhouette: Multi-Armed Bandit Algorithms Library (MAB) :cop:

armature-ai

AI Optimization Framework powered by evolutionary algorithms, multi-armed bandits, and agent societies

polis-protocol

Markdown protocol for multi-vendor AI agent teams — capability cards, bandit routing, and lessons that compound.

arcus-cli

Adaptive routing, a quality gate, and correctness-aware caching on top of Virginia Tech ARC's LLM API - local CLI, your own key, nothing leaves your machine.

mab-algorithm

Code template for multi-armed bandit algorithm

smpybandits

🔬 Research Framework for Single and Multi-Players 🎰 Multi-Arms Bandits (MAB) Algorithms, implementing all the state-of-the-art algorithms for single-player (UCB, KL-UCB, Thompson...) and multi-player (MusicalChair, MEGA, rhoRand, MCTop/RandTopM etc).. Available on PyPI: https://pypi.org/project/SMPyBandits/ and documentation on

ab

A/B Testing Framework for Python (with optional Multi-armed bandit implementation)

banditry

Python library for contextual bandits and Bayesian optimization over mixed design spaces, featuring GP-UCB, Thompson sampling, Gaussian processes, MCMC, and evolutionary acquisition optimization.

slots

A multi-armed bandit library for Python

kvfleet

KV-cache-aware intelligent routing for self-hosted and hybrid LLM fleets. Route requests using model quality, latency, cost, policy, and live GPU state.

improveai

Contextual Multi-Armed Bandit Platform for Scoring, Ranking & Decisions

koth

King of the Hill: performant A/B/n testing beyond the basics. Allocation for multi-armed bandits with correlated arms, using the top-k contenders.

hydraboost

Gradient boosting that picks its own weak learner type each round. Competitive with tuned XGBoost, LightGBM and CatBoost with zero tuning.

contextual-bandits-algos

A comprehensive Python library implementing a variety of contextual and non-contextual multi-armed bandit algorithms—including LinUCB, Epsilon-Greedy, Upper Confidence Bound (UCB), Thompson Sampling, KernelUCB, NeuralLinearBandit, and DecisionTreeBandit—designed for reinforcement learning applications