Bit-TTT is a high-performance AI engine implementing 1.58-bit quantization and Test-Time Training (TTT) entirely in Rust. Unlike typical PyTorch-based workflows, it provides an end-to-end ecosystem—handling tokenization, training, and streaming inference without any Python dependencies. Built on the Candle framework, it ensures memory safety and utilizes CUDA/AVX acceleration, offering a lightweight, standalone alternative for local LLM development.