Open managed agents for swappable harnesses and LLMs
Open managed agents keep the durable agent outside any one model vendor. The Spinup Agent has a real environment, tools, secrets, skills, and state. The harness inside it is swappable. The model path the harness talks to can change as better options appear.
What Open Means Here
Open posture, not just an open-source label
Open managed agents are about where the agent lives and what stays portable when the ecosystem moves. Open-source projects sit inside this picture, but the load-bearing word is portable.
Agent ownership stays with the workspace
The agent definition is not owned by one model vendor. The Spinup Agent is workspace-owned: stable ID, owner, lifecycle, configuration, and recorded run history.
Harness-flexible
The harness inside the environment is swappable. OpenClaw and Hermes today. Claude Code, Codex CLI, OpenCode, PI, Deep Agents, and NanoClaw on the way. The agent does not change when the harness does.
Model-flexible
The model path the harness talks to can change. Closed models, hosted open-weight routes, and self-hosted paths all live inside the same agent definition.
State stays above the provider
Skills, secrets, environment, and run history travel with the agent. State is canonical above the harness and the model, not scattered inside one provider's stack.
Why This Exists
Claude Code exists and OpenCode still matters
The same logic shows up one layer higher. Provider-native managed agents exist. Open managed agents still matter.
When a model vendor ships a flagship coding agent, the flagship is the fastest path for teams that want to be on that vendor. It is also, by design, the wrong shape for teams that want to keep model choice open. That is the gap OpenCode fills against Claude Code: same category, open posture, no provider lock at the harness layer.
The same pattern applies to managed agents. A model vendor shipping a managed-agent platform is the fastest path for teams that want that vendor's stack. It is also, by design, the wrong shape for teams that want the managed-agent category without binding the agent definition, the memory, the tools, and the model choice to one provider.
Open managed agents are the second shape. The Spinup Agent is the durable object. The harness is swappable. The model path can change. The agent stays the same across each layer underneath.
Model Paths
The model layer Spinup is built to support
Each new harness brings its own model paths into the runtime. The agent definition does not have to move when the model path does. Use these as posture, not as live support claims, except where the harness is already live.
Claude
Claude through Claude-native harnesses (Claude Code, NanoClaw) once they land inside the runtime.
OpenAI
OpenAI through Codex CLI and other OpenAI-shaped harnesses as harness coverage expands.
Open-weight routes
Open-weight model paths through OpenRouter and Ollama: Kimi, DeepSeek, GLM, OpenAI OSS models, and others as the open-weight ecosystem moves.
Honest about what runs today
OpenClaw and Hermes today are the supported harnesses. Their model paths run on the runtime now. The rest is posture, not promise.
Harness Set
The harness layer stays open
Each harness is one execution engine inside an isolated agent environment. The agent definition sits above all of them.
OpenClaw
OpenClaw runs today on the local backend inside an isolated microVM, with projected secrets, managed configuration, and persistent state.
Hermes
Hermes runs today through `/exec` inside an isolated microVM, with a persistent Python environment and the same plumbing as OpenClaw.
More harnesses on the way
OpenCode, Claude Code, Codex CLI, PI, Deep Agents, and NanoClaw are on the way. Each one slots in as another harness inside the same agent runtime.
The pattern is the contract
Future harnesses join the same model. The Spinup Agent stays the part that lasts.
FAQ
Common questions about open managed agents
What are open managed agents?+
Open managed agents are managed agents that do not bind the agent definition, environment, harness, or model path to one model vendor. The Spinup Agent stays workspace-owned. The harness inside its environment is swappable. The model the harness talks to can change. Skills, secrets, and run history travel with the agent across every layer underneath.
Does open mean open source?+
Open here means open posture, not necessarily open source. Spinup is built to support open-source harnesses (OpenClaw, OpenCode, Hermes), open-weight model paths through OpenRouter and Ollama, and provider-neutral skills and secrets. Some parts of Spinup are not open source. The point of "open" is that the agent is not owned by one model vendor.
Can I use Claude in an open managed-agent runtime?+
Yes. Claude is one of the model paths Spinup is built to support as harness coverage expands. The difference from a Claude-native managed agent is that the agent definition does not bind to Claude. The same Spinup Agent can run a Claude-based harness, an OpenAI-based harness, or an open-weight harness as the ecosystem changes.
Can I use open-weight models?+
Spinup should make room for Claude, OpenAI, Kimi, DeepSeek, GLM, OpenAI OSS models, and other hosted or open-weight paths as harness support expands. Today, OpenClaw and Hermes are the supported harnesses. Each new harness brings its own model paths into the runtime. The agent definition does not have to move when the model path does.
How does Spinup keep agent state portable?+
Skills, secrets, environment, and run history are bound to the Spinup Agent, not to the harness or the model. Change the harness underneath: the agent stays. Change the model the harness talks to: the agent stays. The agent is the canonical object the workspace owns.
Related
Where open managed agents fit
Read the category page next, then see how a specific harness fits inside this runtime, or how the open path compares to the Claude-native one.
Early access
Managed agents that stay open.
One Spinup Agent above harnesses, models, environments, skills, secrets, and state. Run an open managed agent on OpenClaw or Hermes today.