TinyMaix is a tiny inference library for microcontrollers (TinyML) that converts and runs neural network models in INT8/FP32/FP16 formats. The core library is under 400 lines of code with a .text section under 3KB, enabling even ATmega328-class devices to run MNIST inference. It supports multiple architecture accelerations including ARM SIMD/NEON, RISC-V, CSKYV2, and X86 SSE2, and provides simple APIs to load and run models on resource-constrained hardware.