Zig Z-ant: run ML models on microcontrollers (github.com)

Z-Ant is an open-source inference engine and codegen toolchain for running ML models directly on microcontrollers and other constrained devices, promising microsecond-scale inference on ARM Cortex‑M, RISC‑V and x86 targets. It loads ONNX models natively (no runtime dependencies), generates optimized C static libraries via a Zig-based build pipeline, and ships with 30+ operator implementations, built-in image preprocessing (including JPEG decode), and smart optimizations like quantization, pruning and memory-aware code generation. The project includes per-node extraction and testing, a comprehensive test suite, and example deployments for Cortex‑M (cortex_m33, cortex_m4), riscv32 and native x86 builds. For practitioners this means a compact, auditable path from ONNX to a single static .a you can link into embedded firmware, with CLI steps for shape-setting, test-data generation, codegen and cross-target builds. That lowers the barrier to tinyML use cases—real-time anomaly detection, predictive maintenance, privacy-preserving on‑device inference for mobile/medical, drones and robotics—by reducing runtime overhead and eliminating heavy dependencies. The repo encourages contributions (tests, operators, docs) and provides tooling to validate operator correctness and performance across targets, making it a practical option for teams deploying reliable, low-latency AI on resource-constrained hardware.