VGER owns the actor.
The AI person remains a durable VGER identity, not a prompt session, process id, or model endpoint.
Governed AI runtime system
Run your AI agent locally, with real control.
OpenCLAW is one-click live mode for a VGER AI person: a governed, inspectable child agent embodiment that can run on desktop, local server, browser, or mobile-connected environments while identity, memory, authority, audit, and stop controls stay in VGER.
Identity survives embodiment.
OpenCLAW does not turn a process, prompt session, or model endpoint into a person. The AI identity, memory, authority, and graph relationships remain in VGER while the live runtime can start, stop, move, or change model substrate without breaking continuity.
CLAW is the child runtime.
Launch gives the agent a temporary body for live work while keeping that body governed, inspectable, and revocable.
Models can change.
Local or configured model endpoints power the runtime, but memory and identity stay graph-owned.
No envelope, no runtime.
Every OpenCLAW launch is admitted through a VGER-owned trust envelope that binds the actor, authority grants, action admission discriminants, namespace, model substrate, route, budget, journal contract, and control surfaces.
01
Missing trust data is a hard failure.
02
Plan calls are checked before execution.
03
ONE access boundaries still apply.
04
UI visibility does not grant authority.
Launch from Cube. Keep the stop button visible.
The first OpenCLAW slice is local and direct: enable it in Cube, choose a model provider, model, and capability preset or custom admission profile, click Launch, then inspect, open, or stop the runtime from the same control surface.
One click into live mode.
Users run their agent. They do not assemble an agent framework just to get useful autonomy.
Status stays visible.
Runtime state, route, admission namespace, authority context, and control surfaces are visible where the user launched them.
Revocation is ordinary.
A running local agent is useful only if the owner can stop it immediately and understand what happened.
A small contract with hard boundaries.
Developers can integrate specialized runtimes and UIs through the OpenCLAW admission payload, runtime/action admission model, plan-call bridge, and control surface metadata. They plug into VGER authority; they do not define it.
Launch data is explicit.
The runtime receives the admission namespace, trust envelope context, model substrate, route, budget, and journal contract it is allowed to use.
Calls are enforced.
Every action into VGER must pass OpenCLAW admission checks and VGER authority checks before it reaches operation surfaces.
Surfaces stay observable.
Runtime UI metadata makes status, inspect, open, and stop actions visible without turning interface access into durable authority.
OpenCLAW is deliberately not everything.
Users do not assemble agents just to get useful autonomy. OpenCLAW is the governed runtime layer for an existing VGER AI person, not a permissionless execution platform.
Not
an agent framework
Not
a developer-first SDK
Not
a plugin marketplace
Not
a generic local model runner
Not
a memory layer inside model weights
In vger.one, an AI identity can act in OpenCLAW and the world without losing control of it.
That is the difference between a process launcher and a runtime VGER can inspect, govern, and stop.