Agent harness

Multi-agent orchestration

The harness is the execution loop that turns a model into an agent. It exposes tools to the model, executes tool calls, manages context and memory, applies approval rules for sensitive actions, and delegates to sub-agents. The model reasons, the harness acts: Hermes and OpenCode are two harnesses.

Key points: Execution loop · Tools and context · Memory management · Approvals · Sub-agents.

The execution loop

The harness is the loop that turns a model into an agent. It exposes tools to the model, executes tool calls, feeds the results back, and continues until the task is complete.

Tools, context and memory

The harness manages the context window and memory, selects the relevant tools, and maintains task state between steps.

Approvals

Sensitive actions, such as writing files or executing commands, go through an approval rule. The agent stays under control.

Model and harness

The model reasons, the harness acts. Hermes and OpenCode are two harnesses built on this principle, one for system action, the other for development.