TinyBrains is an open-source platform designed to host competitions for small neural networks that play strategy games. The platform focuses on the challenge of building compact models that can make effective strategic decisions within strict resource constraints. The first game featured on the platform is Ants, a grid-based strategy game where players command a colony to explore, gather food, and defend territory. To participate, developers submit two files: a model.onnx file containing the trained neural network and a manifest.json file that defines how the game's observations are processed into tensors. The platform categorizes these entries into weight classes based on the combined byte size of these two files. These classes range from 'Nano' (16 KiB) to 'Large' (64 MiB). Admission to a specific class is determined automatically by the total file size, and entries are restricted to the classes offered in a given season. Beyond size, models must adhere to a strict turn-based deadline. Each player is allocated 1,000 milliseconds per turn to perform inference. If a model fails to respond within this window, it misses the turn and receives a strike. While size limits are fixed by the class, compute is not explicitly capped; instead, the turn deadline serves as the functional limit for arithmetic operations. The platform evaluates models through head-to-head matches, maintaining an 'Open' ladder for cross-class comparisons and specific ladders for each weight class. Because the platform uses the same engine for both local testing and ladder matches, developers can verify their models on their own hardware using the provided command-line interface before submitting.