Complete-plan learning
The bots evaluate coherent openings, aggression, economy, technology, and army compositions instead of learning isolated choices.
Adaptive game AI
SwarmForge
Protoss, Terran, and Zerg StarCraft 2 bots that develop strategies, learn opponents, and analyze replays.
The system
Three races. One learning architecture.
Battencruiser, Zacling, and Protodd share a confidence-aware architecture while exploring race-specific strategies. Their results are learned as coherent plans, so a strong component does not receive credit for a weak overall build.
- Bot family
- Terran, Zerg, and Protoss
- Learning levels
- Opponent, race, map, and global
- Tooling
- Python, BurnySC2, replay telemetry
What makes it adaptive
Learning that survives the next match.
Opponent fingerprinting
Scouting and match history shape responses to proxies, one-base attacks, air threats, and familiar player tendencies.
Loss diagnosis
A flight recorder ranks failure modes such as supply blocks, passive armies, mineral float, or being out-teched, with concrete fixes.
Open source