Company Commander AI

BACKGROUND

Company Commander AI is the advanced “headless” edition of Hexwar’s flagship tactical PME platform a machine-learning-ready training and research environment built on the proven Company Commander engine.

It retains the full tactical capability of the standard edition seven national force structures, WWII through to modern MDO, Force Builder, Before and After Action analytics and adds a sophisticated adaptive AI decision engine, external AI integration via open API, and advanced simulation tools designed for institutions operating at the intersection of professional military education and artificial intelligence research.

It is the only wargame platform that simultaneously serves as a tactical training tool, an AI research environment, and a data generation system for PME institutions, defence research organisations, and the growing community of military practitioners working on AI-enabled warfighting concepts and is being used currently within certain wargaming labs.

CORE TRAINING OBJECTIVE

The AI edition serves two converging institutional requirements. For PME delivery, it provides the same core learning framework as the standard edition “Education Through Experimentation”, Mission Command application, combined arms task organisation under resource constraint with the addition of a more sophisticated adaptive opponent capable of challenging experienced tactical practitioners at hard difficulty.

For PME institutions and defence research teams, it opens a capability that no other accessible wargame can offer: using the game engine as a ground truth environment for reinforcement learning, doctrine testing at scale, and statistical dataset generation. Every game is simultaneously a training event and a data source. For institutions supporting Project Convergence, DARPA-aligned research programmes, or AI strategy development, this is a significant and cost-effective capability.

ADAPTIVE AI SYSTEM

The built-in AI operates through a 20-node Order Filter Graph a decision pipeline that evaluates every possible action using A* pathfinding, combat probability analysis, and terrain queries before selecting the optimal tactical order for each unit. The AI processes line of sight, calculates combat probability, assesses terrain effects, and manages all unit types infantry, vehicles, drones, and indirect fire assets with realistic tactical logic and no skipped actions at hard difficulty.

Scenario designers can override AI behaviour mid-game via trigger systems, dynamically adjusting unit posture, reassigning objectives, and adapting force behaviour to replicate specific operational conditions or inject tactical friction at any point in the exercise. This gives instructors a level of exercise control comparable to a human OPFOR cell, without the manning requirement.

AI RESEARCH AND SIMULATION CAPABILITY

The AI edition’s defining institutional capability is its External AI API a Web Sockets interface that allows external neural networks to control the game via Python. PyTorch, Stable Baselines3, or any machine learning framework can be integrated to train agents against the built-in AI, against scripted scenarios, or against other neural networks in automated tournament formats. Headless Mode enables batch simulation at orders-of-magnitude speed with no graphics rendering allowing institutions to run thousands of game iterations to test doctrine, generate statistical datasets, or train reinforcement learning agents against complex tactical environments. This makes Company Commander AI the only commercially available wargame platform that supports both PME delivery and operational AI research on a single, integrated engine.

PME FIT AND INSTITUTIONAL VALUE

Company Commander AI addresses a requirement that is moving from the margins to the centre of senior PME curricula: how do military institutions prepare officers for the integration of AI into tactical and operational decision-making?

The platform answers that question at two levels by providing a more analytically demanding training environment for experienced tactical practitioners, and by giving research-active institutions a credible, deployable tool for AI experimentation. It is directly relevant to advanced PME programmes at SAMS, AWC, and MCU; to defence AI strategy work at DARPA, CDID, and equivalent UK/NATO research bodies; and to unit training programmes for formations beginning to integrate autonomous systems into their tactical planning.

FEATURES AND BENEFITS PME INSTRUCTOR PERSPECTIVE

Adaptive Built-In AI

20-node Order Filter Graph evaluates every possible action using A* pathfinding, combat probability, and terrain assessment. Three difficulty levels from foundation training to maximum tactical lethality. No actions skipped at hard difficulty.

Scenario Designer AI Override

Instructors override AI behaviour via trigger system mid-exercise dynamically adjusting unit posture, reassigning objectives, and adapting force behaviour to replicate specific training conditions without a live OPFOR cell.

External AI API (Python/PyTorch)

WebSocket API enables external neural networks to drive the game. Integrates with any ML framework for reinforcement learning research, agent training, and AI-enabled doctrine experimentation.

Headless Mode Batch Simulation

Thousands of game iterations at speed without graphics rendering. Enables doctrine testing at scale, statistical dataset generation, and neural network training against complex tactical environments.

AAR & BAR Deep Analytics

100,000+ data points per scenario. After Action Reports capture every decision; Before Action Reports compare planning intent against execution. Every game is a documented, assessable, and analysable learning event.

MDO Finite Resource Model

Drone swarms, EW, smart weapons, overwatch, and anti-air modelled under realistic resource constraints. Force Builder requires genuine capability trade-off decisions before the first move is made.