An agent harness is the operating layer around a model
An agent harness is the software that turns a language model into a useful agent. It selects context, exposes tools, applies permission rules, keeps a session, streams progress, and turns tool results back into the next model step. The model supplies reasoning; the harness supplies the working environment.
For a coding agent, that environment includes a repository, Git state, files, diagnostics, and commands. Without a harness, a model can suggest a patch. With one, it can inspect the relevant code, explain its plan, request the right capability, and leave an auditable trail of what happened.
What a useful harness owns
A practical harness owns the parts that should behave consistently no matter which model or user interface is active. That includes provider configuration, context selection, tool contracts, session recovery, approval policy, and interruption handling.
- Provider routing and credentials
- Workspace context and memory
- Read, write, search, Git, and terminal tools
- Permission checks and approval requests
- Streaming events, cancellation, and session state
Why the distinction matters
When the harness is separate from the editor, model, or hosted service, developers keep more options. You can move from a terminal workflow to a desktop workspace, select a different provider for a difficult task, or add a new tool without rebuilding the agent around a single surface.
Truss is built around this separation. Its clients provide the experience while the shared runtime keeps the agent loop, tools, context, and permission boundaries consistent.
Next steps
Put it to work in your own workspace.
Pick the surface that fits your day, then connect a model you control.
FAQ
Questions developers ask before switching.
Is an agent harness the same as an LLM framework?
Not exactly. An LLM framework may help call models or compose prompts. An agent harness adds the operational layer: tools, context, permissions, sessions, streaming, and recovery.
Do coding agents need an agent harness?
A simple chat assistant does not. A coding agent that needs to inspect a workspace, call tools, manage approvals, and resume work benefits from a dedicated harness.