Xyntetik Runner
A single native binary, written from scratch in C, that runs local GGUF language models on CPU, CUDA or Metal, serves them over OpenAI- and Anthropic-compatible APIs, and can train LoRA adapters directly through the quantized weights it serves.
Xyntetik Runner is a from-scratch C inference engine for local GGUF and LoRA-adapted language models, with no Python, pip or third-party runtime dependency beyond the engine itself; it targets CPU (x86 AVX2/FMA, ARM NEON), CUDA and Metal. It exposes OpenAI Chat Completions and Responses APIs plus Anthropic's Messages API, so it can be used as a drop-in backend for existing tools, and closes a truncated JSON or tool-call document to a schema-valid ending when the token budget runs out rather than leaving it malformed.
Training runs LoRA adapters through the same frozen quantized GGUF weights the engine serves, without a separate full-precision training copy, and every run can be replayed from a recorded, signed transcript for reproducibility. The engine tracks memory needs so a model can share GPU capacity with other work and unload when idle.
The project is bootstrapped and released free under Apache 2.0; consulting and enterprise work fund its development.