As of June 29, 2026, llama.cpp is an open-source C/C++ inference engine that powers most of the local-AI ecosystem, with tools like Ollama and LM Studio built on its ggml tensor library. The project has more than 118,000 GitHub stars, over 20,000 forks, and is released under the permissive MIT license, shipping continuous build-tagged releases with the latest being b9838. A tutorial walks through 12 hands-on steps, from cloning the repository to serving an OpenAI-compatible API, enabling users to run an LLM locally in about 40 minutes. The steps include building llama.cpp from source with CMake, enabling GPU acceleration for CUDA, Metal, Vulkan, and ROCm, downloading a GGUF model, running inference with llama-cli, tuning performance, starting the llama-server, calling the API from curl and Python, converting a Hugging Face model to GGUF, quantizing a model, benchmarking throughput, and running llama.cpp as a background service. Every command was verified against the official ggml-org/llama.cpp repository.