ahwurm/localharness

An open-source, model-agnostic agent harness for local LLMs. Define agents in YAML (tools, memory, deny-first permissions) and run them against any OpenAI-compatible endpoint: vLLM, Ollama, LM Studio, or llama.cpp.