Agent harness: the software around a working AI model

An agent harness is the software around a model that lets it do work. It supplies tools and context, manages the session, and handles the steps between a request and a result. The model can suggest an action; the surrounding software determines how that action is carried out and checked.

These guides are for developers choosing or building that surrounding system. Keep the model, the tools and the record of the work distinct. Check what happens when a tool refuses an action, a session stops halfway through, or another assistant takes over. A reliable loop still needs current decisions and a useful account of what the previous run completed.

fenbs supplies that shared work record rather than running the agent loop. Projects, tasks, Decisions and rules, lessons and Where we left off stay available to connected AI apps, so different development tools can continue from the same agreed state.