gameaikit

Game AI techniques in plain Python, the standard library and nothing else: search, adversarial search, planning from PDDL, fuzzy control, steering, utility choice, perception and belief, an opponent model, a validator for a language model's replies, and the gradient descent that trains the models the Godot half reads.

Each technique is the tier-1 half of a Godot addon of the same kit, held to the same answers by shared test vectors. It runs, and is tested, without an engine.

from gameaikit import techniques

for name, found in techniques.available().items():
    print(f"{name:26} {found.about}")

from gameaikit.utility import aim

head = aim.Zone("head", 2.0, (0.0, 1.65), [(-0.12, 1.5, 0.12, 1.8)], 1.0)
torso = aim.Zone("torso", 1.0, (0.0, 1.175), [(-0.2, 0.85, 0.2, 1.5)], 1.0)
print(aim.choose([head, torso], 0.5).aimed)   # 1: far away, the torso

Your own technique

Declare it in your package's pyproject.toml and techniques.available() lists it beside the kit's:

[project.entry-points."gameaikit.techniques"]
my-search = "my_package.search:MySearch"

A name the kit already uses is refused rather than shadowed.

Licence

Apache-2.0. LICENSE and NOTICE travel with the wheel, in its .dist-info/licenses/.