No user-facing changes. This release swaps the local development LLM backend from Ollama to a local llama.cpp server. Production argument import and semantic search are unchanged — they continue to run on OpenAI exactly as before.