Harness

On its own, a model takes text and returns text. The harness is what turns that text into action: it reads the model’s reply, runs the requested tool (open a file, run a command, search the web), sends the result back to the model and repeats until the task is done.

What a harness usually includes

• A loop that calls the model as many times as the task needs.
• Tools: read and write files, run commands, call APIs.
• Context management: what goes into the context window at each step, what gets summarized and what is left out.
• Permissions: what the agent does on its own and what needs approval.
• Extensions, such as skills and connections to other systems.

Why it matters

Two products running the same model can produce very different results because of the harness. Much of the gap between a chat and an agent that works in your files lives here, not in the model.