Local LLMs on your own Mac or NVIDIA machine with llama.cpp, Ollama or LM Studio, so your documents never leave the building. The model is chosen by measuring a few candidates on your own examples, and you get an OpenAI-compatible API your tools can call.