Spinup Agents
The durable product object you configure and run. Memory, capabilities, secrets, harness choices, and run history live with the agent, above any one runtime environment.
A Spinup Agent is the durable thing you configure and run. It keeps memory, capabilities, secrets, harness choices, and run history above any one runtime environment. Each run gets a fresh projection into the selected harness.
The agent is what you own. The environment is the computer it works on, and the harness is the engine inside that environment. Those two pieces are replaceable. The agent is not.
What lives with the agent
These settings belong above any one harness or runtime instance:
- memory (facts, preferences, decisions, lessons)
- capabilities (skills, MCP servers, tools, plugins, CLIs, runtime requirements)
- secrets attached through the workspace
- enabled harnesses and the selected default
- model choice and other configuration
- run history
When Spinup starts an environment for a run, it renders these settings into a harness-specific projection. When the run ends, the projection can be discarded. Your settings stay on the agent.
How you reach an agent
/exec and the in-product chat surface operate on the agent, not on a harness-local profile. The same agent ID accepts a run through OpenClaw today and a run through Hermes the next minute, with the same memory and the same capabilities available to both.
A run can override the harness or model for one request. The override does not change what is saved with the agent.
What this page does not cover
This page describes what you control. It does not describe how Spinup stores agent state, how saved state versions are tracked, or how proposed changes flow back from a harness. Those details belong in the implementation reference, not in the product model.
Where to go next
- See Environments for how the runtime side fits in.
- See Harnesses for how the same agent runs across different harnesses.
- See Runs for the unit a request is billed and recorded against.
- Manage agents from the terminal with List agents and Manage agents.
- Read agents from your own code with the TypeScript SDK.
- Call the workspace-scoped API directly through Control plane agents.