Harnesses
The engine that runs inside the environment. The same Spinup Agent can be projected into different harnesses without losing memory, skills, or secrets.
A harness is the engine that runs inside the environment. It turns a request into an agent loop: it reads the prompt, calls the model, uses tools, and produces output.
OpenClaw and Hermes are the harnesses supported today. More will follow as the ecosystem evolves.
One agent, any harness
The same Spinup Agent can be projected into different harnesses. You enable the harnesses you want available, choose one as the default, and override per run when you need to compare.
Switching harnesses does not lose:
- memory
- capabilities (skills, MCP servers, tools, plugins, CLIs, runtime requirements)
- secrets
- the agent's name and run history
The harness is replaceable. The agent is not. When you point a request at OpenClaw and the next request at Hermes, the same memory and the same capability settings are available to both. Spinup renders a harness-specific projection for each run so the harness sees its native files, prompts, and configuration.
What the harness owns
The harness owns the loop itself: how it reads instructions, when it calls the model, how it invokes tools, and what it writes to disk during a run. Anything the harness wants Spinup to remember has to come back as a proposed change and earn an accepted receipt.
Where to go next
- See Spinup Agents for what travels with the agent across harnesses.
- See Environments for the runtime the harness lives in.
- See Memory for how the harness proposes changes back to Spinup.
- Enable and switch harnesses from the terminal with Agent harnesses.
- Update enabled harnesses through the workspace API with Control plane agents.
- Target a specific harness on a single request through Status and exec.