Definition: The execution layer surrounding the model that manages Agent execution, tools, memory, restrictions, and the decision-making loop.
Simply put: The LLM itself is only the “brain”; the Harness is what turns that brain into an executable Agent.
Examples:
- OpenCode, which provides the model with various tools.
- A system that takes the model’s decision and executes a Function.
- An environment that sends the result of each Tool Call back to the model.